On the convergent evolution of animal toxins. Conservation of a diad of functional residues in potassium channel-blocking toxins with unrelated structures.

BgK is a K+ channel-blocking toxin from the sea anemone Bunodosoma granulifera It is a 37-residue protein that adopts a novel fold, as determined by NMR and modeling. An alanine-scanning-based analysis revealed the functional importance of five residues, which include a critical lysine and an aromatic residue separated by 6.6 ± 1.0 Å. The same diad is found in the three known homologous toxins from sea anemones. More strikingly, a similar functional diad is present in all K+ channel-blocking toxins from scorpions, although these toxins adopt a distinct scaffold. Moreover, the functional diads of potassium channel-blocking toxins from sea anemone and scorpions superimpose in the three-dimensional structures. Therefore, toxins that have unrelated structures but similar functions possess conserved key functional residues, organized in an identical topology, suggesting a convergent functional evolution for these small proteins.

Functional properties of proteins are frequently associated with a small number of important residues. For example, enzyme activities depend on a few residues that are essential for catalysis. Also, protein-protein recognition processes have been predicted (1) and recently demonstrated (2) to be energetically driven by a small proportion of the residues forming the contacting areas in protein-protein complexes, as identified by x-ray studies (3,4). Among the proteins whose major functions require protein-protein interactions are animal toxins, which bind to various molecular targets, such as receptors or ion channels, using a small number of binding residues (5)(6)(7)(8). As has been shown for enzymes (9), toxins with different architectures are capable of exerting similar functions (10). However, in contrast to enzymes, the molecular basis associated with the conservation of the function in structurally unrelated toxins remains unknown. In this paper, we show that two families of animal toxins with different folding patterns but a comparable capacity to bind to potassium channels include similar functional diads, composed of a critical lysine and an aromatic amino acid separated from each other by 6.6 Ϯ 1.0 Å.

Synthesis of Toxin and Mutants-
The amino acid sequence of BgK 1 was proposed a few years ago (11). However, chemical synthesis attempts, based on these data, systematically failed. The proposed amino acid sequence was therefore questioned, re-examined, and ultimately corrected. 2 The revised amino acid sequence of BgK from Bunodosoma granulifera is: VCRDWFKETACRHAKSLGNCRTSQKYRANCAKTC-ELC. BgK and each alanine-substituted analog were synthesized by solid phase synthesis using an Applied Biosystems model 431A peptide synthesizer, starting from 0.1 mmol of Rink-resin (4-(2Ј,4Ј-dimethoxyphenylhydroxymethylphenoxy resin; 0.48 mmol/g). A 10-fold excess (1 mmol) of Fmoc (N-(9-fluorenyl)methoxycarbonyl)-protected amino acid was used and coupled in N-methylpyrrolidone in the presence of N,NЈdicyclohexylcarbodiimide/1-hydroxybenzotriazole. The following side chain protections were used: t-butyl ether for Ser and Thr; t-butyl ester for Glu and Asp; trityl for Cys, His, Asn, and Gln; 2,2,5,7,8-pentamethylchromane-6-sulfonyl for Arg; and t-butoxycarbonyl for Lys and Trp. The peptide was simultaneously cleaved from the resin and deprotected by treatment (500 mg of resin/50 ml) with 90% trifluoroacetic acid, 5% H 2 O, 5% triisopropylsilane for 1.5 h at room temperature. The resin was filtered, and the solution was precipitated with methyl-t-butylether, washed three times with ether, dissolved in 20% acetic acid, and lyophilized.
The crude peptides (0.1 mg/ml) were oxidized in 50 mM phosphate buffer, pH 7.8, containing 5 ϫ 10 Ϫ3 M reduced and 5 ϫ 10 Ϫ4 M oxidized glutathione. Oxidation was complete after 1 h at room temperature. The solution was then acidified to pH 3.0 with acetic acid and directly applied to a Vydac C18 column (25 ϫ 1 cm) for purification. The peptides were eluted with an acetonitrile gradient containing 0.1% trifluoroacetic acid. The fractions containing the oxidized peptides were analyzed by analytical HPLC. The adequately oxidized components were identified by mass spectrometry. The yields in purified and oxidized toxins ranged from 20 to 40%. Disulfide assignments were achieved by peptide mapping using the Lys-C endoprotease (1:20 by mass) in 20 mM Tricine buffer, containing 1 mM EDTA, pH 7.6, for 7 h at 35°C. Digests were submitted to reverse phase-HPLC on a Vydac C18 column (25 ϫ 0.46 cm), and the resulting fragments were collected, lyophilized, and analyzed by electrospray mass spectrometry. The mass of the fragments unambiguously indicated that the pairings were Cys 2 -Cys 37 , Cys 11 -Cys 30 , and Cys 20 -Cys 34 .
Amino acid analyses were performed on an Applied Biosystems model 130A automatic analyzer, equipped with an on-line model 420A derivatizer, and mass analysis on a Nermag R10 -10 mass spectrometer, coupled to an Analitica of Branford electrospray source. Amino acid compositions and mass analyses agreed with the expected theoretical values.
Peptide concentrations were determined by monitoring absorbances * This work was supported by the Atomic Energy Commission, Direction de Recherches et Etudes Techniques, and the Wellcome Trust. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ‡ Present address: CBS, CNRS-UMR 995, Université de Montpellier, 34060 Montpellier Cedex, France.
¶ To whom correspondence should be addressed.
at 280 mm, using a molar extinction coefficient of 8100, calculated on the basis of the known amino acid content (11). For analogs where Trp or Tyr were deleted, the peptide concentrations were determined on the basis of quantitative amino acid analyses. Circular dichroism spectra were recorded using a Jobin-Yvon CD6 dichrograph, driven by an IBM PC operating with a CD6 data acquisition and manipulation program. Spectra in the range 180 -250 nm were run at 20°C in 5 mM sodium phosphate, pH 7.0, with a 0.1 cm quartz cell and a protein concentration of 1.5-2.0 ϫ 10 Ϫ5 M.
NMR Experiments-Synthetic BgK (9.8 mg) was dissolved in 440 l of solvent, leading to a final concentration of 4.9 mM. Solvents were either a mixture of 400 l of water and 40 l of D 2 O, or 440 l of D 2 O. pH was fixed at 3.7, and chemical shifts were measured relative to TSP-d 4 . All NMR experiments were performed at 600 MHz (AMX600 Brü ker spectrometer) at 20 and 30°C, in order to resolve assignment ambiguities. At each temperature and in each solvent, a DQF-COSY spectrum (12), a nuclear Overhauser spectroscopy spectrum (300 ms mixing time) (13), and a total correlation spectroscopy spectrum (14,15) were recorded. Quadrature detection was performed using the States method (16) in the indirect dimension and using simultaneous mode acquisition in the directly detected dimension. The water signal was suppressed by low power irradiation at all times except during t 1 and t 2 . The spectra were recorded with 128 (t 1 ) ϫ 1024 (t 2 ) (96 ϫ 1024 in D 2 O) and with a sweep width of 7812 Hz (6578 Hz in D 2 O). The 3 J NH-H␣ (respectively 3 J H␣-H␤ ) coupling constants were measured from the high resolution DQF-COSY in H 2 O (respectively D 2 O).
Signal Transformation and Molecular Modeling-All data were transformed using FELIX software (17). Prior to Fourier transformation, spectra were weighted with a 18°shifted sine bell (90°for the spectra to be integrated) and were zero-filled in the t 1 dimension to yield 1000 ϫ 1000 matrices after reduction (1000 ϫ 4000 for the DQF-COSY). Each NOE was integrated with FELIX, using a 2.48 Å distance between the ␦ and the ⑀ protons of Tyr 26 for calibration. A range of Ϯ25% of the distance value was used to define the upper and the lower bounds of the restraints. Structures were obtained by molecular modeling using successively DIANA (18) for preliminary structure calculation and X-PLOR for simulated annealing (19,20). After having solved assignment ambiguities with the help of DIANA, 50 structures were calculated using this software, and the 30 structures with a target function below 10 were further refined by simulated annealing in X-PLOR 3.1 (with files parallhdg.pro and topallhdg.pro). The 15 structures with the lowest energy were kept for analysis. Within these 15 structures, all but three distance violations are no larger than 0.3 Å and all dihedral restraints violations are lower than 5°.
Competition Binding Experiments-Competition experiments were made on rat brain synaptosomes using 125 I-labeled ␣-dendrotoxin, as described previously (21).

RESULTS
Structure Determination-BgK was isolated from B. granulifera (11), and its amino acid sequence (see Fig. 1) has been recently corrected. 2 BgK, like a number of scorpion toxins such as charybdotoxin, (i) binds to potassium channels, (ii) contains 37 residues, and (iii) possesses 6 half-cystines. However, our attempts to align the amino acid sequences of BgK and scorpion toxins have failed, suggesting that BgK adopts a structure that differs from the well-known ␣/␤ fold of scorpion toxins (22). We therefore decided to elucidate its structure by NMR and molecular modeling.
Structural Assignment-Three unique amino acids having specific spin systems were used as starting points for sequential assignments. These are Trp 5 , His 13 , and Gly 18 . The unique tyrosine (Tyr 26 ) was also used to assign the C-terminal part of the amino acid sequence of BgK. As no proline was present in the protein, these four starting points as well as the existence of all but one sequential NH-NH connectivities allowed us to achieve the complete assignment of the amino acid sequence of BgK (Table I).
Experimental Restraints, r.m.s.d., and Energetical Parameters-A total of 767 distance restraints deriving from NOEs (316 intra-residual correlations, 172 sequential, 155 short range (͉i Ϫ j͉ Յ 4), 124 long range (͉i Ϫ j͉ Ͼ 4) and 12 dihedral restraints (7 from angles, 5 from 1 angles) deriving from coupling constants was used. This large set of constraints (21 restraints by residue) yielded a well defined structure, as reflected by the mean r.m.s.d. value of its backbone, which was as low as 0.8 Ϯ 0.2 Å between two structures. The calculated structures were consistent with both experimental data and the standard covalent geometry. The structures had no distance violations larger than 0.32 Å, and only three distance violations were larger than 0.3 Å. All dihedral restraint violations were lower than 5°. The covalent geometry was respected, as revealed by the low ͗r.m.s.d.͘ values of the bond length (0.003 Å) and the valence angles (1.52°).  (47), a chemical shift variation larger than 0.1 from the average value of the ␣-proton for the concerned residue is represented by an arrow, directed upward for a positive difference and downward for a negative difference. A shift increase is expected in ␤-sheet or extended strand, while a helical conformation will yield a decrease of the chemical shift.
Backbone Structure Description-The existence of two helices, running from residues 9 to 16 and residues 24 to 31, was clearly indicated by the presence of numerous short-range NOEs between h␣(i) and hn(i ϩ 3) protons and between h␣(i) and h␤(i ϩ 3) protons ( Fig. 1). No other regular secondary structure emerged from these data. Fig. 2a shows the 15 best structures of BgK, which result from transformation of NMR data and molecular modeling. The two helical stretches are well defined with r.m.s.d. values of 0.4 Ϯ 0.1 Å and 0.5 Ϯ 0.2 Å for the position 9 -16 and 24 -31 stretches, respectively. Though to a lesser extent, the other parts of the toxin structures were also relatively well defined. Panels b and c of Fig. 2 show, respectively, the overall fold of BgK and the spatial organization of its secondary elements. The N-and C-terminal regions are maintained in spatial proximity by the disulfide 2-37, whereas the third disulfide 20 -34 brings the loops 16 -24 and 31-37 inside the center of the molecule, providing the toxin with a globular shape. The 11-30 disulfide, which links the two helices, adopts a unique conformation, which is almost lefthanded (23). It is centrally located in the structure, consistently with a tight organization of a local hydrophobic core. The other disulfide bonds adopt either two conformations (disulfide 20 -34) or no prevailing conformation (disulfide 2-33).
Mutational Analysis-To identify functional residues of BgK, we submitted the toxin to an alanine-scanning experiment, producing all single point variants by chemical synthesis, as described under "Materials and Methods." Competition experiments were performed between the variants and radiolabeled ␣-dendrotoxin on membrane from rat brain synapto-somes. Twenty-five variants have been synthesized and investigated regarding their effects on dendrotoxin binding. Fig. 3 shows five inhibition curves obtained with native BgK and four variants, which display a substantially lower competition activity, as compared with native BgK. Except for the variant T33A, all inhibition curves are parallel to that observed with the native toxin. Similar parallel inhibition curves were obtained for other variants, except K25A, for which no inhibition was observed even in the presence of 3 ϫ 10 Ϫ7 M protein. In the presence of the variant F6A, a 10% increase in binding of labeled dendrotoxin was observed. Although we are not able to explain this observation, as yet, one should recall that different subtypes of potassium channels are present in brain and are composed of heterooligomeric mixtures of different protein subunits. Experiments with subtype-specific antibodies reveal that most of the channels in the brain contain K v1.2 (ϳ80%) or K v1.1 (ϳ50%) subunits (24). ␣-Dendrotoxin blocks K v1.1 and K v1.2 channels with almost equal affinity (IC 50 values ϳ 20 nM), and has greater than 10 times less affinity for other cloned channels (25). Therefore, in view of such complexity, the present competition data have to be considered with caution. From curves similar to those shown in Fig. 3, we deduced IC 50 values for all variants. These values are compiled in Table II. The correlation coefficient for fitting the data points was calculated from the Hill equation, y ϭ R max /[1 ϩ (X/IC 50 ) P ]. As can be seen from Table II, a value close to 1 was obtained in all cases. Moreover, preliminary competition experiments performed with cloned K v1.2 channels and a number of the variants exhibiting a lower inhibitory capacity, as compared to native BgK, nicely paral- leled those reported in Table II. 3 Therefore, we anticipate that data obtained with brain synaptosomes mostly reflect the functional importance of residues implicated in the capacity of BgK to bind to K v1.2 potassium channels. Nevertheless, in order not to overinterpret our data, we considered only the relative inhibitory capacity of the variants, without any attempt to deduce their binding affinities. Competition experiments, shown in Table II, revealed that mutation of lysine 25 into alanine is associated with the largest affinity decrease. This mutation, however, caused no significant change in the secondary structure of the toxin, as inferred from the circular dichroism spectra of the variant which is quite similar to the spectrum of the native toxin (Fig. 4). We conclude, therefore, that the low competition ability of this mutant is not due to distortions in the toxin structure but to the absence of the lysine side chain and possibly to the loss of its positive charge. Upon mutations of three other residues, Phe 6 , Tyr 26 , and Thr 33 , BgK was a less potent inhibitor, since its competition capacity decreased by factors of 46, 38, and 19, respectively. However, while the circular dichroism spectra of the first two variants were indistinguishable from that of the native toxin, the spectrum of the T33A variant showed some distortions (data not shown), which could not be readily interpreted but which might reflect some structural perturbations. Additionally, the inhibition curve with the T33A variant (see Fig. 3) was not quite parallel to those observed with the native BgK or other variants. In addition, synthesis of this variant was particularly difficult to achieve, leading to a relatively low yield of recovery. Therefore, all these observations suggest that introduction of an alanine at position 33 may be associated with structural perturbations, which might account for the substantial decrease in inhibitory activity of BgK. As a result, one cannot safely propose that the side chain of residue 33 is implicated in the binding of the toxin to the BgK target. Therefore, in the absence of further data, only the side chains of Lys 25 , Phe 6 , and Tyr 26 are concluded to be involved in the functional site of BgK. Mutations at two positions, Ser 23 and His 13 , also induced approximately 8-and 6-fold decrease in competition capacity. Although relatively low, these values might also reflect involvement of these residues in the binding site of BgK. Mutations at the other positions either had no effect on the affinity of BgK or caused changes in inhibitory capacity that are lower than 3-fold. These side chains are not considered, therefore, as actors in the recognition capacity of BgK for dendrotoxin-sensitive binding sites. In summary, alanine-scanning-based experiments indicated that three, and perhaps five, residues of BgK belong to the surface by which the toxin interacts with the channels. a number of toxins, found in venoms of (i) sea anemones such as BgK (11) or ShK (28), (ii) scorpions like charybdotoxin (29) or noxiustoxin (30), and (iii) snakes like dendrotoxin (27), bind to dendrotoxin-sensitive K ϩ channels (31,32). However, it is unclear as to whether these proteins share any structural or chemical elements that could account for their common capacity to recognize similar channels.
In this paper, we first provided evidence that BgK from the sea anemone B. granulifera is structurally unrelated to scorpion toxins, although these two groups of toxins share a similar number of residues, the same number of disulfides, and comparable functional properties. The present structural analysis revealed that the fold adopted by BgK contains two nearly perpendicular stretches of helices, with no additional canonical secondary structures (Fig. 2). The globular architecture of the toxin is stabilized by the three disulfides, one of them linking the two helices. This structure is similar to that of ShK, as indicated by a paper that was published during the review of our manuscript (33). ShK, from the Stichodactyla helianthus sea anemone, is a 35-residue toxin that also binds to potassium channels (28). As in BgK, the structure of ShK includes two helices running from residues 14 -19 and 21-24. However, these two stretches are not located at identical corresponding positions in both toxins. In fact, the absence of four consecutive residues in the amino acid sequence of ShK, as compared to BgK (residues 12-15 using BgK numbering), leads to a different organization of the N-terminal part of the toxin. More precisely, the N-terminal part of BgK is successively composed of an extended strand (1)(2)(3)(4)(5)(6)(7)(8), an helix 1 (9 -16), and a loop (17)(18)(19)(20)(21)(22)(23), whereas in ShK, it is composed of an extended strand (1-8), a loop (9 -13), and helix 1 (14 -19). The structures of the C-terminal parts of the two toxins are more similar in the two toxins. Therefore, BgK and ShK adopt the same type of fold with, however, a number of marked structural differences located mostly in the N-terminal regions. Quite distinctly, charybdotoxin (ChTX), a typical scorpion toxin that blocks potassium channels (29), adopts a different fold with a short helix linked by two disulfides to a three-stranded ␤-sheet (22). No ␤-sheet structure was found in BgK.
Although BgK and ChTX have different structures, they both inhibit the binding of labeled dendrotoxin to potassium chan-nels (11,31), suggesting that they bind to regions of the channels that are recognized by dendrotoxin. In agreement with this view, previous observations reported that the sites recognized by ChTX (34,35) and dendrotoxin (36,37) commonly include the very conserved P-region of the pore of the channels. The channel region that is recognized by BgK is unknown; however, the toxin binds to several subtypes of K v1 channels (i.e. K v1.1 , K v1.2 , and K v1.3 ) with almost equal affinity, 2 suggest-

TABLE II
Effect of introduction of an alanine residue at 25 individual positions of BgK on the ability of the toxin to inhibit binding of 125 I-labeled ␣dendrotoxin to membranes from rat brain synaptosomes All residues except half-cystines, alanines, and the unique glycine have been mutated. Mutation at Asp 4 yielded no refolded compound. The S.E. values for IC 50 values are derived from independent experiments (n ϭ 5 for the wild type toxin and three for each variant). The displacement curves were fitted with a sigmoid curve for a single binding site. The correlation coefficients for fitting the data points were calculated from the Hill equation, y ϭ R max /[1 ϩ (X/IC 50 ) P ]. Due to various reasons (see text) the decrease in inhibitory activity observed upon mutation at Thr 33 uniquely was not considered to reflect the functional importance of the threonine residue. WT, wild type. ing that the target of BgK also includes a highly conserved region of the channels, possibly the P-region. In the absence of further data on channel regions that are recognized by these toxins, one way to understand the molecular basis associated with their common capacity to inhibit the binding of dendrotoxin consists in comparing the sites by which these toxins recognize their targets. With a view toward identifying the functionally important residues of BgK, we submitted the toxin to an alanine-scanning experiment and compared the ability of all the synthetic variants to inhibit the binding of dendrotoxin membranes from rat brain synaptosomes. Twenty-five positions out of 37 have been modified. Therefore, if one excepts the six half-cystines, which are likely to play a structural role, nearly 80% of the positions of BgK have been individually explored regarding their possible involvement in the binding of the toxin to dendrotoxinsensitive sites in rat brain synaptosomes. Our data showed that introduction of an alanine at five positions, i.e. Lys 25 , Phe 6 , and Tyr 26 , and, to a lesser extent, at His 13 and Ser 23 , caused a decrease in the ability of the toxin to compete with dendrotoxin for its specific binding sites, without changing the secondary structure of the toxin. Moreover, preliminary and unpublished data performed with cloned K v1.2 channels and a number of BgK variants parallel those obtained with rat brain synaptosomes, strongly supporting the view that the residues Lys 25 , Phe 6 , Tyr 26 , and perhaps His 13 and Ser 23 , are involved in the functional site of BgK. Of these residues, however, mutation at Lys 25 most dramatically affected the capacity of BgK to inhibit dendrotoxin binding, suggesting that this particular lysine is the major binding actor of BgK. This finding agrees with recent observations made with ShK, a homologous K ϩ channel-blocking toxin from the sea anemone S. helianthus whose lysine 22, which corresponds to Lys 25 in BgK, plays a critical binding role toward the same channels (38). Previously, ChTX was submitted to site-directed mutagenesis, and the residues by which the toxin binds to the voltage-sensitive Shaker K ϩ channel were identified (8,39). Clearly, Lys 27 was the most critical residue, with four neighboring amino acids (Tyr 36 , Met 29 , Asn 30 , and Arg 34 ) whose mutation also affected the affinity of ChTX to the channel. Thus, the functional residues of ChTX form a homogeneous area located on the flat ␤-sheet face that is exposed to solvent (8,39) and that covers approximately 200 Å 2 (18 ϫ 12 Å). Although BgK has no ␤-sheet structure, it nevertheless possesses a flat surface of similar size (21 ϫ 9 Å), which is formed by the edge of the 9 -16 helix and its two flanking loops. Strikingly, this surface harbors the functional residues of BgK (see Fig. 5A). Therefore, BgK and ChTX possess comparable flat surfaces, which include a similar small number of energetically important residues, among which a lysine is the major binding actor (Fig. 5B).
If one assumes that these lysines play a similar binding role in BgK and ChTX, their superimposition can be readily associated with superimposition of Tyr 26 in BgK with Tyr 36 in ChTX, two aromatic residues that play an important binding function in the toxins (Fig. 6, top). The distances that separate the C␣ of lysines from the center of the benzene rings of the tyrosines are 6.6 Ϯ 1.0 Å. The common capacity of the two toxins to recognize potassium channels is therefore associated with the conservation of a similar diad of functional residues, a lysine and a close aromatic residue. The other functional residues are different in the two toxins and do not form any evident superimposable pattern. Remarkably, however, if one rotates ChTX 90°around a central axis localized near the crucial lysine, Phe 6 of BgK superimposes with Tyr 36 in ChTX (Fig. 6,  bottom). Thus, by virtue of the four-fold symmetrical organization of the channel (40), Phe 6 in BgK is likely to interact with the same aromatic binding site as Tyr 26 , but on another monomer of the channel.
Similar diads are present in toxins that are homologous to BgK and ChTX. Thus, the diad Lys 25 -Tyr 26 is conserved in the three known K ϩ channel-blocking toxins from sea anemones (Fig. 7). The situation is slightly more complex with K ϩ channel-blocking scorpion toxins. Thirteen K ϩ channel blocking scorpion toxins are presently known. They have been previously divided into four subfamilies (41). Three of these subfamilies possess the same Lys 27 -Tyr 36 diad as in charybdotoxin. In contrast, all toxins forming the fourth subfamily, i.e. the two kaliotoxins and the three agitoxins (AgTX 1-3 ), have a threonine at position 36. Strikingly, all these toxins uniquely possess a phenylalanine at position 24 or 25, located on the exposed face of the ␤-sheet. Mutational studies of a member of this family (AgTX 2 ) showed that this phenylalanine is involved in the binding to the Shaker potassium channel (42). Interestingly, the diad Phe 25 -Lys 27 in AgTX 2 can be superimposed on the diad FIG. 5. A, stereoview of a structure of BgK and its functional residues. Lys 25 , whose mutation caused the larger decrease in inhibitory capacity, is shown in red. Tyr 25 and Phe 6 , whose mutations caused a decrease in inhibitory activity by factors higher than 10, are shown in orange. His 13 and Ser 23 , whose mutations caused the lowest effect, are in yellow. Note that these residues occupy a flat surface defined by an edge of the longest helix and the two sandwiching loops. The homogeneous surface that is covered by these residues is also shown in B, where the toxin atoms are represented in CPK. The color code of the functional residues is the same as in A.
Lys 27 -Tyr 36 in ChTX provided the ␤-sheets of the two toxins form a 90°angle. We suggest, therefore, that Tyr 36 in toxins of three subgroups of scorpion toxins and Phe 25 in the remaining toxins occupy the same binding site on the four-fold symmetrical channel, but on different monomers. Thus, the common capacity of sea anemone and scorpion toxins to recognize K ϩ channels is associated with the conservation of a functional diad, composed of an essential lysine assisted by a 6.6 Ϯ 1.0 Å distant aromatic residue whose precise nature (Tyr or Phe) and location may differ from one toxin to another.
Why is such a diad conserved among two families of functionally similar but structurally unrelated proteins? In scorpion toxins, the positively ammonium group of the lysine of the diad may mimic K ϩ ions entering the pore, occluding the ion pathway (39,43,44). A similar role may be associated with the functional lysine of the toxins from sea anemones. The role played by the aromatic residue of the diad is less obvious. Recent bimutational analyses (42) showed that Phe 25 in AgTX 2 mainly interacts in the Shaker with the hydrophobic Met 448 , a position that is conserved or conservatively mutated (Ile, Val) in other K ϩ channels (45). Then, the conserved aromatic residues in the diads provide the toxin with two unique features. First, they offer the possibility for local hydrophobic interactions to take place between toxins and channels. Second, since their rings are somewhat parallel to the plane defined by the flat surface of the toxin, they allow the crucial lysine to protrude outside the toxin surface, thus being in an appropriate position to plug into the pore. Evidently, residues other than those of the diads also play some functional role in the toxins. These additional residues clearly differ from one toxin to another. Therefore, we suggest that the diads constitute a con-served functional core around which less conserved residues provide the different toxins with a proper binding specificity, which varies from one toxin to another. A similar situation that has also been found with snake curaremimetic toxins (5,46).
In conclusion, two families of toxins with distinct folding patterns display a conserved functional diad. Each diad contains a critical lysine which should possess environmental and/or structural features that other lysines do not share and which allow it to exert a predominating recognition role toward potassium channels. Its location on a flat surface, its protruding position and the constant close proximity of an aromatic ring probably contribute to give a unique role to the functional lysine. Whether similar criteria are shared by other structurally unrelated but functionally similar toxins remains to be investigated. FIG. 6. A, superimposition of the functional diad Lys 27 /Tyr 36 (in pink) of the scorpion charybdotoxin whose backbone is in red, with either the functional diad Lys 25 (violet)/Tyr 26 (yellow) (top) or Lys 25 (violet)/Phe 6 (orange) (bottom) of BgK whose backbone is in cyan. Note that for both superimpositions to be achieved, the structure of the scorpion toxin has to be rotated by 90°around an axis centered around the common lysine. ChTX coordinates come from the Protein Data Bank.
FIG. 7. Amino acid sequences of the three known homologous toxins from sea anemone that recognize potassium channels. BgK and ShK were, respectively, isolated from B. granulifera and S. helianthus (11,28). In both names, K stands for potassium, the targeted ion channels, and the first two letters correspond to the initials of the producing species. AsKS, also called kaliseptine, has been found in Anemonia sulcata extracts (51).