Candoxin, a novel toxin from Bungarus candidus, is a reversible antagonist of muscle (alphabetagammadelta ) but a poorly reversible antagonist of neuronal alpha 7 nicotinic acetylcholine receptors.

In contrast to most short and long chain curaremimetic neurotoxins that produce virtually irreversible neuromuscular blockade in isolated nerve-muscle preparations, candoxin, a novel three-finger toxin from the Malayan krait Bungarus candidus, produced postjunctional neuromuscular blockade that was readily and completely reversible. Nanomolar concentrations of candoxin (IC(50) = approximately 10 nm) also blocked acetylcholine-evoked currents in oocyte-expressed rat muscle (alphabetagammadelta) nicotinic acetylcholine receptors in a reversible manner. In contrast, it produced a poorly reversible block (IC(50) = approximately 50 nm) of rat neuronal alpha7 receptors, clearly showing diverse functional profiles for the two nicotinic receptor subsets. Interestingly, candoxin lacks the helix-like segment cyclized by the fifth disulfide bridge at the tip of the middle loop of long chain neurotoxins, reported to be critical for binding to alpha7 receptors. However, its solution NMR structure showed the presence of some functionally invariant residues involved in the interaction of both short and long chain neurotoxins to muscle (alphabetagammadelta) and long chain neurotoxins to alpha7 receptors. Candoxin is therefore a novel toxin that shares a common scaffold with long chain alpha-neurotoxins but possibly utilizes additional functional determinants that assist in recognizing neuronal alpha7 receptors.

adjacent loops that emerge from a small globular core, which is the location of the four conserved disulfide bridges (3)(4)(5)(6)(7)(8)(9). Other members of this family include -bungarotoxins, which recognize neuronal nicotinic receptors (10), muscarinic toxins with selectivity toward distinct types of muscarinic receptors (11), fasciculins that inhibit acetylcholinesterase (12), calciseptins that block the L-type calcium channels (13,14), cardiotoxins (cytotoxins) that exert their toxicity by forming pores in cell membranes (15), and dendroaspins, which are antagonists of various cell adhesion processes (16). Despite their common structural fold and comparable affinity for the Torpedo and muscle (␣␤␥␦) nAChRs, ␣-neurotoxins are classified as short chain neurotoxins (e.g. erabutoxin-b (Laticauda semifasciata)) that have 60 -62 residues and four conserved disulfide bonds and long chain neurotoxins (e.g. ␣-bungarotoxin (Bungarus multicinctus); ␣-cobratoxin (Naja kaouthia)) with 66 -75 residues and five disulfide bonds (3). The additional disulfide bridge in long chain ␣-neurotoxins, as well in the neuronal -bungarotoxin (B. multicinctus) is located in the middle loop (loop II) (3,8,9). This fifth bridge, which cyclizes a helix-like conformation at the tip of loop II, has been reported to be crucial for long chain ␣and -neurotoxins to bind to ␣7 and ␣3␤2 neuronal nAChRs, respectively (17,18), but not to Torpedo or muscle (␣␤␥␦) nAChRs. Consequently, short chain ␣-neurotoxins that lack this fifth disulfide bridge bind to neuronal nAChRs with about 3 orders of magnitude lower affinity than long chain neurotoxins (17,19). In addition, detailed site-directed mutagenesis studies have shown that in both long and short chain ␣-neurotoxins, functionally invariant residues in loops I, II, and III participate in binding to nAChRs, with the role of particular loops varying according to the type of neurotoxin (short or long) and the type of nAChR (muscle or neuronal ␣7) (19 -23).
The weak toxins, which constitute another class of threefinger toxins, consist of 62-68 amino acid residues and five disulfide bridges. However, unlike in the long chain ␣and -neurotoxins, the fifth disulfide bridge in weak toxins is located in loop I (N-terminal loop) (24). Toxins belonging to this class were first isolated from Naja melanoleuca venom (25) and were also referred to as the melanoleuca type (26,27) or the miscellaneous type of toxins (28). They are typically characterized by a lower order of toxicity (LD 50 varying from ϳ5 to 80 mg/kg) as opposed to prototype ␣-neurotoxins (LD 50 varying from ϳ 0.04 to 0.3 mg/kg) (29). Apart from toxicity studies, weak toxins have been poorly investigated in terms of their * 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.
FIG. 1. Isolation, purification, and assessment of homogeneity of candoxin. A, gel filtration chromatography of crude B. candidus venom on a Superdex 30 column with Tris-hydrochloride buffer (50 mM; pH 7.5; flow rate ϭ 1.5 ml/min) as eluent. Elution of proteins was monitored at 280 nm (solid line), 215 nm (dashed and dotted line), and 254 nm (dotted line). The fractions indicated by the horizontal bar were pooled and subjected to reverse-phase HPLC. B, reverse-phase HPLC of pooled gel filtrations on a Jupiter C18 column using a linear gradient (20 -45% over 80 min) (dotted line) of buffer B (80% acetonitrile in 0.1% trifluoroacetic acid) at a flow rate of 2 ml/min. The peak indicated by the arrow was rechromatographed on a shallower gradient. C, rechromatogram obtained by reverse-phase HPLC using a shallow gradient (28 -36% over 50 min) (dashed line) of buffer B (80% acetonitrile in 0.1% trifluoroacetic acid). Elution was monitored at 280 nm (black line) and 215 nm (gray line). The resulting single protein peak (indicated by the arrow) was named candoxin. D, analytical reverse-phase HPLC (on a Sephadex C18 column using a linear gradient of buffer B (dashed line) at a flow rate of 200 l/min) chromatogram of purified candoxin (CDx) and of a 1:1 mixture of purified candoxin (CDx) and the major ␣-neurotoxin (␣-NTx) present in B. candidus venom. Elution was monitored at 280 nm (solid line) and 215 nm (dotted function or molecular targets. The three-fingered fold is also adopted by proteins from nonvenom sources like the Ly-6 family of cell surface accessory proteins expressed on immune system cells (30 -32). Interestingly, a murine gene lynx 1, which is highly expressed in the brain, has been found to encode for a three-finger protein that is a novel modulator of neuronal (␣7 and ␣3␤2) nAChRs in vitro (33). This raises the possibility that snake toxins, particularly the weak toxins that have an additional disulfide bond in loop I as do Lynx 1 and the Ly-6 family of proteins (30 -33), may be evolutionarily related to an endogenous ligand for neuronal nAChRs.
Recently, Utkin et al. (24) have reported that a weak toxin (WTX) isolated from N. kaouthia venom produced an almost irreversible inhibition of acetylcholine (ACh)-evoked currents at the muscle and human or rat ␣7 nAChRs in micromolar concentrations. This is significantly less potent than the typical long chain ␣-neurotoxins that inhibit both ␣7 and muscle (␣␤␥␦) nAChRs in nanomolar concentrations (17). We now report the isolation, purification, and pharmacological and electrophysiological characterization, as well as some structural data of a novel toxin, candoxin, from the venom of the Malayan krait Bungarus candidus. Nanomolar concentrations of candoxin produced a readily reversible block of muscle (␣␤␥␦) nAChRs, whereas it produced a poorly reversible block of neuronal ␣7 receptors, clearly showing diverse functional properties toward the two subsets of nicotinic receptors. In light of the scientific and clinical significance of neuronal nAChRs (34), candoxin could play an important role as a biological marker of ␣7 receptors.

EXPERIMENTAL PROCEDURES
Materials-Lyophilized B. candidus venom was obtained from Venom Supplies (Tanunda, Australia). Prepacked chromatography columns were purchased from Amersham Biosciences. All drugs and chemicals were purchased from Sigma with the exception of the following, which were obtained from the sources indicated: reagents for Nterminal sequencing (Applied Biosystems, Foster City, CA), acetonitrile (Fisher), and trifluoroacetic acid (Fluka Chemika-Biochemika, Buchs, Switzerland). HPLC grade water was obtained by using a Milli-Q purification system (Millipore). ␣-Cobratoxin, ␣-bungarotoxin, and erabutoxin-b were from Latoxan (Valence, France).
Purification of Candoxin-B. candidus venom was fractionated based on the molecular weight of the components on a Superdex 30 fast performance liquid chromatography column (1.6 ϫ 60 cm) equilibrated with Tris-hydrochloride buffer (50 mM; pH 7.5). The fraction containing candoxin was further purified on a reverse-phase Jupiter C18 (0.21 ϫ 25 cm) column using a Vision Biocad Work station (Bio-Rad). The column was equilibrated with 0.1% trifluoroacetic acid, and the proteins were eluted with a linear gradient (20 -45% over 80 min) of 80% acetonitrile in 0.1% trifluoroacetic acid (buffer B). The HPLC peak containing candoxin was rechromatographed by reverse-phase HPLC using a shallower gradient of buffer B (28 -36% over 50 min). Elution of proteins was monitored at 280 and 215 nm.
Electrospray Ionization Mass Spectrometry-Candoxin was subjected to electrospray ionization (ESI) mass spectrometry using a PerkinElmer Life Sciences Sciex API 300 triple quadrupole instrument equipped with an ion spray interface. The ion spray and orifice voltages were set to 4600 and 30 V, respectively. Nitrogen was used as a curtain gas with a flow rate of 0.6 liter/min, and compressed air was used as a nebulizer gas. The sample was infused by flow injection at a flow rate of 50 l/min using Shimadzu 10 AD pumps as the solvent delivery system.
Matrix-assisted Laser Desorption Ionization-Time of Flight Mass Spectrometry-Matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) mass spectrometry was performed on a Voyager DE-STR Biospectrometry work station (Applied Biosystems). The ma-trix used was saturated with (10 mg/ml) sinapinic acid (3,5-dimethoxy-4-hydroxycinnamic acid) in 1:1 acetonitrile:water containing 0.3% trifluoroacetic acid. Candoxin (5 pmol in 1 l) was spotted onto a stainless steel sample plate with 1 l of matrix solution and dried off. The accelerating voltage was set at 25,000 V, and the grid and guide wire voltages were set at 93.0% and 0.3%, respectively. Molecular ions were generated using a nitrogen laser (wavelength 337 nm) at an intensity of 1800 -2200. Extraction of ions was delayed by 800 ns. The spectrum was calibrated using external standards.
Capillary Electrophoresis-Capillary electrophoresis was performed on a BioFocus3000 system (Bio-Rad). Candoxin (50 g) was injected to a 25 m ϫ 17 cm coated capillary using a pressure mode (5 p.s.i./s) and run in 0.1 M phosphate buffer (pH 2.5) under 18.00 kV from positive to negative at 20°C for 7 min. Migration was monitored at 200 nm.
Determination of the N-terminal Amino Acid Sequence-Candoxin was resuspended in 100 l of denaturant buffer (6.0 M guanidinium hydrochloride, 0.13 M Tris, 1 mM EDTA, pH 8.0) containing 0.07 M ␤-mercaptoethanol. The solution was heated at 37°C for 2 h. Subsequently, 1.5-fold molar excess (over sulfhydryl groups) of 4-vinylpyridine was added and incubated at room temperature. After 2 h, the sample was desalted by reverse-phase HPLC. N-terminal sequencing of the native and pyridylethylated protein was done by automated Edman degradation using a PerkinElmer Life Sciences 494 pulsed liquid phase protein sequencer (Procise) with an on-line 785A phenylthiohydantoinderivative analyzer.
Chick Biventer Cervicis Muscle-The pair of chick biventer cervicis muscles (CBCM) were isolated from 7-10-day-old chicks (35) and mounted in 8-ml organ baths containing Kreb's solution of the following composition 118 mM NaCl, 4.8 mM KCl, 1.2 mM KH 2 PO 4 , 2.5 mM CaCl 2 , 25 mM NaHCO 3 , 2.4 mM MgSO 4 , and 11 mM D-(ϩ) glucose, with a resting tension of ϳ1 g. The solution was maintained at 34°C and aerated with 5% carbon dioxide in oxygen. Motor responses of the muscle were evoked by stimulating the motor nerve supramaximally by electrical field stimulation (7-10 V, 0.1 ms, 0.2 Hz) using a Grass stimulator and recorded in a Mac Lab system 8 TM (AD Instruments, Sydney, NSW Australia) via a force displacement transducer (model FT03). Direct muscle stimulation was achieved by electrical field stimulation (20 -30 V, 1 ms, 0.2 Hz) in the presence of d-tubocurarine (10 M) to block neuromuscular transmission. Submaximal contractures to exogenously applied ACh (300 M for 30 s), carbachol (8 M for 90 s), and potassium chloride (30 mM for 60 s) were obtained in the absence of electrical field stimulation prior to the addition of the candoxin (0.1-100 g/ml; 0.0136 -13.6 M) and at the end of the experiment. The neuromuscular blockade produced by candoxin was compared with that produced by 0.005-20 g/ml of erabutoxin-b (36) (0.733-2.9 M), ␣-bungarotoxin (37) (0.625-2.5 M), and ␣-cobratoxin (38) (0.626 -2.55 M). The neuromuscular block is expressed as a percentage of control twitch height after 15 min of exposure of the CBCM to the respective toxins. The recovery of the CBCM from complete neuromuscular blockade produced by candoxin or 90% blockade produced by erabutoxin-b, ␣-bungarotoxin, or ␣-cobratoxin, was assessed by washing the respective toxin by bath overflow with fresh Kreb's solution until maximal recovery. The effects of the anticholinesterase, neostigmine (0.1-3 M) on the reversal of neuromuscular blockade were also studied.
Analysis of Primary and Tertiary Structure of Candoxin-The amino acid sequence of candoxin was subjected to a similarity search using BLAST (39), and multiple sequence alignment was done using CLUST-ALW (40). The solution NMR structure of candoxin (Protein Data Bank 2 S. Nirthanan and R. M. Kini, unpublished observations. line). E, capillary electrophoresis of candoxin. The sample was injected using pressure mode 10 p.s.i./s, and electrophoresis runs were carried out using a coated capillary (17 cm ϫ 25 m) from positive to negative polarities at 12 kV, with 100 mM phosphate buffer (pH 2.5) at 18°C for 20 min. F, electrospray ionization mass spectrum of candoxin. The spectrum shows a series of multiply charged ions, corresponding to a single, homogenous peptide with a molecular weight of 7334.6. Inset, reconstructed spectrum. G, MALDI-TOF mass spectrum of candoxin. The spectrum is the average of 233 scans. The peaks depicting m/z 7335.62, 3668.47, and 2445.81 represent (M ϩ H) ϩ , (M ϩ 2H) 2ϩ , and (M ϩ 3H) 3ϩ ionization states of candoxin, respectively. accession code 1JGK) 3 was analyzed with respect to the 2.0-Å crystal structure of erabutoxin-a (Protein Data Bank accession code 5EBX) (41) and the 2.4-Å crystal structure of ␣-cobratoxin (Protein Data Bank accession code 2CTX) (42) deposited in the Protein Data Bank (43). The structures were annotated and visualized using WeblabViewerLite, version 3.2 (Molecular Simulations Inc.).
Oocyte Preparation and cDNA Injection-Xenopus oocytes were isolated and prepared as described previously (44). The oocytes were injected intranuclearly with expression vectors containing the various cDNAs (2 ng) and incubated for 2-3 days at 18°C in Barth's solution (88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO 3 , 10 mM HEPES, 0.82 mM MgSO 4 ⅐7H 2 O, 0.33 mM Ca(NO 3 ) 2 ⅐4H 2 O, and 0.41 mM CaCl 2 ⅐6H 2 O). The pH was adjusted to 7.4 with NaOH. All of the subunits were injected in equal concentrations. To minimize contamination, the medium was supplemented with 20 g/ml of kanamycin, 100 units/ml penicillin, and 100 g/ml streptomycin.
Electrophysiological Recording-Electrophysiological recordings were performed with a two-electrode voltage clamp (GeneClamp amplifier; Axon Instruments, Foster City, CA) as described previously (17,45). The cells were held at Ϫ100 mV and continuously superfused with OR2 (oocyte ringer) medium (82.5 mM NaCl, 2.5 mM KCl, 5 mM HEPES, pH 7.4, adjusted with NaOH) with 2.5 mM Ca 2ϩ during the recording. The flow rate was ϳ6 ml/min, and the volume of the chamber was less than 100 l. To ensure equilibrium of the blockade, the oocytes were exposed to candoxin for 30 min in all of the experiments. To prevent adsorption of the toxin on the plastic, bovine serum albumin was added to the perfusion and candoxin media at a concentration of 20 mg/ml.
Electrophysiological Data Analysis-Concentration-response curves were adjusted using the empirical Hill equation.
where Y is the fraction of remaining current, IC 50 is concentration of half-inhibition, nH is the apparent cooperativity, and x is antagonist concentration. For muscle (␣␤␥␦) receptors, the inhibition curve was best fitted with a two-component Hill equation. When two Hill equations were employed (see Fig. 4B), the sum of two identical equations was computed.

RESULTS
Isolation and Purification of Candoxin-Candoxin was purified to homogeneity by consecutive gel filtration and reversephase HPLC. B. candidus venom was fractionated on a Superdex 30 gel filtration column, and the fractions constituting peak 4 (Fig. 1A, indicated by horizontal bar) were found to produce neuromuscular blockade in the isolated CBCM. The proteins in peak 4 were further fractionated on a reverse-phase HPLC Jupiter C18 column as shown in Fig. 1B. The protein peak identified was rechromatographed by reverse-phase HPLC using a shallower gradient (Fig. 1C, arrow). The single protein peak thus obtained by rechromatography was named candoxin. To ensure the absence of contaminants, especially by other ␣-neurotoxin(s) present in the venom, the purified sample of candoxin was subjected to several sensitive assays and found to be homogenous by analytical reverse-phase HPLC (Fig. 1D), capillary electrophoresis (Fig. 1E), ESI mass spectrometry (Fig. 1F), and MALDI-TOF mass spectrometry (Fig.  1G). Moreover, the elution peaks of candoxin (molecular weight, 7334.6) and the ␣-bungarotoxin-like protein (molecular weight, 7984.4) were widely separated in the chromatogram obtained by analytical reverse-phase HPLC (Fig. 1D). Finally, during the process of amino acid sequencing by automated Edman degradation, there was no evidence for the presence of any peptide contaminants, even when up to 9 nmol of candoxin was loaded on the sequencer. Candoxin has a molecular mass of 7334.67 Ϯ 0.35 as determined by ESI mass spectrometry (Fig.  1F) and reconfirmed by MALDI-TOF (7334.69 Ϯ 0.26) (Fig.  1G). It constitutes about 1-2% of the crude venom.
Comparison of the Amino Acid Sequence of Candoxin-We were able to unequivocally identify all of the residues and determine the complete amino acid sequence of both native (blank cycles where cysteine residues are found) and pyridylethylated candoxin samples. Candoxin has 66 amino acid residues including 10 cysteine residues. The calculated mass of candoxin is 7344.4, and with the expected five disulfide bridges, the calculated mass coincides well with the estimated molecular masses of 7334.67 Ϯ 0.35 (ESI mass spectrometry) and 7334.69 Ϯ 0.26 (MALDI-TOF mass spectrometry). The amino acid sequence of candoxin (Fig. 2) is deposited in the SWISS-PROT protein data base (accession number P81783). The disulfide linkages, established by the observation of long range H␤-H␤ nuclear Overhauser effects in the nuclear Overhauser effect spectrum in NMR studies, showed the presence of five disulfide bridges, of which those seen between Cys 3 and Cys 26 , Cys 19 and Cys 43 , Cys 47 and Cys 59 , and Cys 60 and Cys 65 were homologous to the four conserved disulfide bridges found in other members of the three-finger toxin family (Fig. 2). 3,4 The fifth disulfide bridge in candoxin is located at the tip of loop I (Cys 6 -Cys 11 ) instead of in loop II as found in other ␣-neurotoxins (46). Candoxin shares ϳ30 -40% identity with short and long chain ␣-neurotoxins and -bungarotoxins, as well as other three-finger toxins ( Fig. 2A). The four conserved disulfide bridges contribute significantly (ϳ20%) to the similarity. Candoxin is identical but for one or two residues to two long neurotoxin homologues deduced from cDNA sequences from B. multicinctus. Other weak toxins, including WTX (24) and ␥-bungarotoxin (47), showed ϳ40 -45% identity with the exception of a neurotoxin homologue (EMBL accession number CAC50565) deduced from its cDNA sequence from a coral snake Micrurus corallinus (Elapidae) that showed significant (56%) identity. Interestingly, candoxin showed only 29% identity to bucandin, a three-finger toxin structurally related to weak toxins that was purified from the same venom (48).
Neuromuscular Blockade Produced by Candoxin in Vitro-Mice injected (intraperitoneally) with candoxin showed flaccid paralysis of the hind limbs (data not shown). Accordingly, we screened candoxin for biological activity on isolated nerve muscle preparations. Candoxin produced a rapid, concentration-dependent blockade of the twitch responses of the CBCM to indirect (nerve) stimulation as well as a complete block of the responses to exogenously applied nicotinic agonists, ACh and carbachol (data not shown). Neither the twitch responses elicited by direct muscle stimulation nor the responses to exogenously applied KCl (50 mM) were affected by candoxin (up to 200 g/ml; 27.2 M) (data not shown). The neuromuscular blockade was sustained for over 90 min without spontaneous reversal, following which the twitch responses evoked by indirect stimulation were rapidly and completely restored by washing the organ bath with fresh Kreb's solution (Fig. 3, A and B). In another series of experiments, 0.1, 1, and 3 M neostigmine produced complete reversal of the neuromuscular blockade produced by candoxin in 12 Ϯ 1.2, 7 Ϯ 0.9, and 3 Ϯ 0.3 min, respectively (data not shown). In contrast, erabutoxin-b, ␣-bungarotoxin, and ␣-cobratoxin produced neuromuscular blockade that was ϳ6 -10-fold more potent than that produced by candoxin (IC 50 ϭ ϳ1.5 M) (Fig. 3C) but that was virtually irreversible even after prolonged washing for 180 min. Moreover, the addition of the neostigmine (up to 100 M) did not reverse the neuromuscular blockade significantly once it had progressed to complete block (data not shown).
Electrophysiological Studies-Electrophysiological experiments on various subtypes of nAChRs expressed in Xenopus oocytes were designed to further elucidate the molecular tar-get(s) of candoxin. Firstly, oocytes expressing the rat muscle (␣␤␥␦)receptors were challenged with different concentrations of candoxin. Application of candoxin alone produced no detectable current, whereas it strongly inhibited ACh-evoked currents in the rat muscle (␣␤␥␦) receptors. As shown in Fig. 4A, incubation with 100 nM candoxin (30 min) inhibited about two-thirds of the ACh-evoked current. Recovery from candoxininduced blockade was rapid and complete following a 10-min wash. The inhibition dose response over a broad range of candoxin concentrations revealed a half-inhibition (IC 50 ) of ϳ10 nM

FIG. 2. Comparison of the amino acid sequence of candoxin with the sequences of neurotoxins (A) and some weak toxins from
Elapidae venom (B). The cysteine residues are shaded. The disulfide linkages and segments contributing to the three loops are also shown. The percentage of identity of the respective toxins to candoxin is indicated. The sequence data were obtained from either the Swiss Prot (Swiss Institute for Bioinformatics) or EMBL (European Bioinformatics Institute) protein data bases. The toxin names are followed by species names and references in parentheses.

FIG. 3. Reversible neuromuscular blockade produced by candoxin in vitro.
A, segment of tracing showing the blockade of nerve-evoked twitches of the chick isolated biventer cervicis muscle produced by candoxin (100 g/ml). The neuromuscular blockade was sustained for ϳ90 min, and the toxin was washed out of the bath at the point indicated. B, the time course of reversal of the neuromuscular blockade produced by various concentrations of candoxin. Ⅺ, 10 g/ml; E, 30 g/ml; छ, 100 g/ml. The recovery is calculated as a percentage of the control twitch responses. C, concentration-response curves for the neuromuscular blockade produced by candoxin (Ⅺ), ␣-bungarotoxin (छ), ␣-cobratoxin (‚), and erabutoxin-b (E). The block is calculated as a percentage of the control twitch responses of the muscle to supramaximal nerve stimulation. Each data point is the mean Ϯ S.E. of at least six experiments. (Fig. 4B, open squares). However, attempts to describe these data points with a single Hill equation yielded a very low Hill coefficient and a poor fit quality (Fig. 4B, continuous gray line), and best fit was obtained with the sum of two Hill equations with high (2.2 nM, nH ϭ 1.6) and low affinity (98 nM, nH ϭ 1.4) (Fig. 4B, continuous dark line). The high and low affinity components contributed almost equivalently to the inhibition (46 and 54%, respectively). The plateau phase observed between the high and low affinity curves in Fig. 4B is suggestive of the existence of two binding sites. In agreement with this hypothesis, the blockade produced by conotoxin MI (49) measured under the same experimental conditions also yielded a doseresponse inhibition curve that presents a plateau phase (data not shown). When candoxin was applied to oocytes expressing the major brain nAChR (␣4␤2), no detectable effects were observed (data not shown). In contrast, a strong inhibition of the ACh-evoked current was measured in oocytes expressing the rat ␣7 receptor, the pattern of sensitivity to ␣7 receptors resembling that observed for other snake neurotoxins such as ␣-bungarotoxin and ␣-cobratoxin. As shown in Fig. 4C, incubation with 300 nM candoxin resulted in a significant inhibition of the ACh-evoked current. Surprisingly however, no recovery of the response could be detected after a 10-min wash, and the absence of recovery was noted even up to 6 h following the exposure to candoxin. Partial recovery of the response was observed after 24 h. Determination of the inhibition doseresponse curve yielded a half-inhibition (IC 50 ) at ϳ50 nM (Fig.  4D, open squares). In contrast to the muscle (␣␤␥␦) receptors, the inhibition curve was adequately described by a single Hill equation with a Hill coefficient of 0.85 (Fig. 4D, continuous line). DISCUSSION The fifth disulfide bridge in candoxin is located at the tip of loop I instead of in loop II as found in other long chain ␣-neurotoxins (46) (Fig. 5, B and C). This disulfide motif of candoxin places it among members of the family of weak toxins (8,24). At present, about 25 amino acid sequences of weak toxins have been identified either from their cDNA sequences or by their isolation from venoms. Weak toxins have been isolated only from snakes of the Elapidae family, mostly from the Naja (cobras) and Bungarus (kraits) spp. but as well as from Dendroaspis jamesoni (mamba) and M. corallinus (coral snake). Although, weak toxins derive their name by virtue of their low toxicity, their lethality (LD 50 ) has been shown to vary widely (5-80 mg/kg) (27,29). For instance, ␥-bungarotoxin (B. multicinctus) that is structurally related to weak toxins has a LD 50 of 0.15 mg/kg, comparable with ␣-neurotoxins (47), whereas weak toxin WTX (N. kaouthia) did not kill mice at concentrations of up to 2 mg/kg (24). Candoxin showed a LD 50 of 0.83 mg/kg (by intravenous injection) in mice that was only ϳ 6 -8fold less potent than the lethality of curaremimetic ␣-neurotoxins.
Pharmacological Studies-Our pharmacological studies showed that candoxin produced concentration-dependent blockade of the responses of isolated nerve-muscle preparations (CBCM) to both nerve stimulation and exogenously applied nicotinic agonists (ACh and carbachol), indicating a postjunctional site of action. Because it did not block responses to direct muscle stimulation by KCl, the postjunctional activity cannot be attributed to an effect on muscle contractility but rather to its blockade of nAChRs of the neuromuscular junction. These effects closely resembled the neuromuscular blockade produced by erabutoxin-b, ␣-bungarotoxin, and ␣-cobratoxin, well known for their selective and high affinity binding to postjunctional nAChRs (50). However, the neuromuscular blockade produced by candoxin was ϳ6 -10-fold less potent than that produced in vitro by short and long neurotoxins, but more significantly, it was readily and completely reversed by washing or by the addition of the anticholinesterase neostigmine, which is in sharp contrast to the virtually irreversible neuromuscular blockade produced by most ␣-neurotoxins (50). Neostigmine overcomes the neuromuscular blockade produced by candoxin as a consequence of acetylcholinesterase inhibition and acetylcholine preservation at the synapse, presumably resulting in competitive displacement of candoxin. The block produced by ␣-neurotoxins, however, is not reversed as a result of their poorly reversible binding to muscle nAChRs (51). This reversible nature of the candoxin-induced neuromuscular blockade was also demonstrated in the mouse isolated hemi-diaphragm as well in the tibialis anterior muscle of anesthetized rats. 5 Utkin et al. (24) have drawn attention to the importance of ensuring that ␣-neurotoxins are not present as trace contaminants in the sample of weak toxins being subjected to functional studies. Accordingly, we have taken elaborate measures to ensure the homogeneity of candoxin (see "Results"). In addition, our in vitro and in vivo pharmacological studies consistently showed that the neuromuscular blockade produced by candoxin was completely and rapidly reversible, which would not have been the case if an ␣-bungarotoxin-like protein contributed wholly or in part to the blockade.
Electrophysiological Studies-Our electrophysiology data on the block of oocyte-expressed neuromuscular junction (muscle ␣␤␥␦) receptors by candoxin are in agreement with that obtained in vitro on the isolated chick muscle (CBCM). Specifically, rapid and complete recovery was observed even following complete block of ACh-evoked currents by high concentrations of candoxin. Despite the disparity in experimental design, a clear difference was, however, observed between the half-inhibition in these two preparations, suggesting a preferential sensitivity of candoxin to murine receptors. Because the usual prey of the Malayan krait B. candidus is rodents and reptiles (52), it is possible that its venom components may be well selected for these species. A similar phenomenon has been reported previously for ␣-bungarotoxin, which displayed a higher affinity for rat rather than chick nAChRs (53). Because it has been clearly established that ␣-bungarotoxin binds to the N-terminal domain of the nAChR subunits (54), these data suggest that determinant residues for toxin binding may be different between the subunits of the chick and rat nAChRs.
As shown in Fig. 4B, the inhibition dose-response curve for the block of muscle (␣␤␥␦) receptors by candoxin is best fitted by the sum of two Hill equations with a high (2.2 nM, nH ϭ 1.6) and low (98 nM, nH ϭ 1.4) affinity. Interestingly, the high affinity component displays an IC 50 close to that previously reported at the rat (␣␤␥␦) receptor for ␣-bungarotoxin (IC 50 4.9 nM, nH ϭ 1.1) (55). Because these data have been obtained using a large number of cells from different batches (n ϭ 28), it could be argued that the high and low affinity may relate to interexperiment variation rather than the interaction between candoxin and the receptors. Because the same phenomenon was, however, observed on a single cell in which subsequent candoxin exposures were effectuated (data not shown), this dual affinity component must be attributed to properties of the receptors. This observation is supported by the comparison of the fits obtained with a single or dual Hill equation (Fig. 4B, continuous gray line versus continuous dark line). The latter reveals the presence of a plateau phase between the high and low affinity dose-response inhibition curves of candoxin. Additional evidence for the existence of a high and a low affinity site that can be functionally revealed was provided by the observation of a similar plateau phase in the dose-response inhibition curve of conotoxin MI that is highly selective for the ␣/␦ binding site in mouse muscle nAChRs (56) (data not shown).
Challenging oocytes expressing the neuronal ␣4␤2 or ␣7 nAChRs with candoxin unveiled the paradoxical nature of this toxin. Although, as expected from previous studies carried out with ␣-bungarotoxin, no blockade was observed at ␣4␤2 receptors (57), a marked inhibition of the ACh-evoked current was observed at the ␣7 receptors (53,58). The rat ␣7 receptors displayed an IC 50 of ϳ 50 nM to candoxin with a Hill coefficient of 0.85. Our data also revealed a higher apparent affinity of candoxin for the neuronal ␣7 nAChRs than ␣-conotoxin-ImI (IC 50 of ϳ 220 nM, nH ϭ 0.89) (55). Erabutoxin-a (17,59) and the weak toxin (WTX) from N. kaouthia (24) showed very poor affinity in micromolar inhibitory concentrations for ␣7 receptors (Fig. 4D). By comparison, ␣-bungarotoxin blockade at these receptors (IC 50 0.52 nM, nH ϭ 1.9) is markedly more pronounced than candoxin with a difference of almost 2 orders of magnitude and a higher Hill coefficient (55). Nonetheless, the blockade of nAChRs by candoxin and ␣-bungarotoxin revealed interesting differences. The most important was the full reversibility of the candoxininduced block at the muscle (␣␤␥␦) receptors that profoundly contrasts with the extremely slow time course of recovery at the ␣7 receptor (Fig. 4, A and C), whereas ␣-bungarotoxininduced block is almost irreversible at both receptor subsets (17,55). To rule out the possibility of the ␣7 blockade being 5  attributed, wholly or in part, to a putative contaminant, protection experiments using a competitive antagonist were further effectuated. Application of the ␣7-specific, competitive antagonist methyllycaconitine (1 M) (60) was found to protect ␣7 receptors from candoxin blockade (data not shown), thereby illustrating that these two compounds must interact at the same binding site on the receptor. This suggests that the poorly reversible blockade of ␣7 receptors seen with candoxin must be attributed to candoxin and not to the presence of a putative contaminant. It should also be remembered that although there appear to be discrepancies between the apparent affinity of these toxins for muscle or ␣7 receptors and their time course of recovery from receptor blockade, functional investigations (such as by electrophysiology) only measure the concentration of the antagonist required for half-blockade of the receptors and may not correlate with the actual affinity of the antagonist for the receptor as conventionally determined by binding studies.
Comparison of the Structure of Candoxin with Neurotoxins-Notwithstanding their classification as short and long chain neurotoxins, both types of curaremimetic neurotoxins bind with high affinity to the Torpedo and muscle ␣␤␥␦ nAChR. In contrast, only long chain neurotoxins are able to recognize the neuronal ␣7 nAChR with high affinity (17,19,23). Clearly, the two families of curaremimetic toxins, which share many structural similarities, are not functionally homogenous. Despite a common structural fold, candoxin, which blocks both the muscle (␣␤␥␦) and ␣7 receptors at relatively low nanomolar concentrations, showed distinct differences from long chain neurotoxins with respect to structure and function. The first loop of candoxin was found to be longer than that of ␣-cobratoxin, and it also lacked the long C-terminal tail that is a characteristic feature of most long chain neurotoxins ( Figs. 2A and 5, B and C). In these aspects, it appeared more similar to erabutoxin-a (Fig. 5, A and C). The middle loop of ␣-cobratoxin differed markedly from that of candoxin, because of the presence of a small helix-like segment cyclized by the fifth disulfide bridge, which previous studies (17,23) have shown to be critical for long chain neurotoxins to block ␣7 receptors in nanomolar concentrations. The structure of the ␣7 receptor-specific conotoxin-ImI also consists of a functionally important helical scaffold resembling the helix-like tip of the middle loop of long chain neurotoxins (61). Short chain neurotoxins (17), WTX, a weak toxin from N. kaouthia (24), and an atypical long chain neurotoxin (neurotoxin-b) from Laticauda colubrina (17), which lack this fifth disulfide bridge and helix-like segment in their middle loop, have weak affinity, in micromolar concentrations at best, for the ␣7 receptor. Interestingly, although candoxin lacks this helix-like segment in its middle loop, it blocks ␣7 receptors in low nanomolar concentrations.
Putative Determinants for the Muscle ␣␤␥␦ Receptor-Short and long chain neurotoxins recognize the Torpedo (or muscle ␣␤␥␦) receptor by a cluster of positively charged and aromatic residues that constitute a common binding core. These are, in erabutoxin-a and ␣-cobratoxin, 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 . In addition, each toxin also utilizes specific residues for receptor recognition. In erabutoxin-a, these include His 6 , Gln 7 , Ser 8 , Gln 10 , Gly 34 , Ile 36 , and Glu 38 (20,21). Specific determinants for ␣-cobratoxin include Arg 36 and Phe 65 (19,22). Of these critical residues for the muscle ␣␤␥␦ receptor, Trp 25 , Arg 33 and Arg 36 (in ␣-cobratoxin) and Glu 38 and Gly 34 (in erabutoxin-a) are present at homologous positions in candoxin ( Figs. 2A and 6A). The substitution of the critical residue Lys 27 /Lys 23 (in erabutoxin-a/␣-cobra-toxin, respectively) by Glu 29 in candoxin has also been reported in some other neurotoxins that retain full neurotoxicity (3). The root mean square deviation between Trp 25 /Arg 33 / Arg 36 in ␣-cobratoxin and Trp 31 /Arg 35 /Arg 38 in candoxin was 0.96 Å, whereas the root mean square deviation between Lys 27 /Trp 29 /Arg 33 in erabutoxin-a and Glu 29 /Trp 31 /Arg 35 in candoxin was 0.89 Å. This is comparable with the root mean square deviation (0.98 Å) between Lys 27 /Trp 29 /Arg 33 in erabutoxin-a and Lys 23 /Trp 25 /Arg 33 in ␣-cobratoxin, indicating a remarkable similarity in the spatial disposition of these critical residues for the muscle receptor in candoxin, erabutoxin-a, and ␣-cobratoxin.  ␣7 receptor (C and D). The solution NMR structure of candoxin (Protein Data Bank accession code 1JGK) 3 was analyzed with respect to the 2.0-Å crystal structure of erabutoxin-a (Protein Data Bank accession code 5EBX) and the 2.4-Å crystal structure of ␣-cobratoxin (Protein Data Bank accession code 2CTX). A, putative residues by which candoxin may interact with the Torpedo receptor, based on data for erabutoxin-a (21, 69) and ␣-cobratoxin (22). A space-filling (Corey-Pauling-Koltun) molecular representation of the concave surface of candoxin is shown. Residues identified as critical in erabutoxin-a and/or ␣-cobratoxin are showed in red. Gly 36 has been reported to be involved but not crucial for the binding of erabutoxin-a (21). The substitution of the critical Lys 27 (in erabutoxin-a) by Glu 29 in candoxin has also been reported in other curaremimetic neurotoxins with high affinity for the Torpedo receptor (3). B, a side view representing the ␣-carbon backbone structure of candoxin showing the complete side chains of the putative, functionally important residues for interaction with the Torpedo receptor facing the concave surface. C, putative residues by which candoxin may interact with the neuronal ␣7 receptor, based on data for ␣-cobratoxin (23). A space-filling (CPK) molecular representation of the concave surface of candoxin is shown, and residues identified as critical in ␣-cobratoxin are shown in green. D, a side view representing the ␣-carbon backbone structure of candoxin showing the complete side chains of the putative, functionally important residues for interaction with the ␣7 receptor facing the concave surface.
Putative Determinants for the Neuronal ␣7 Receptor-Mutagenesis studies on ␣-cobratoxin (22,23) revealed that it recognizes both the muscle ␣␤␥␦ and ␣7 receptors, with six common residues (Trp 25 , Asp 27 , Phe 29 , Arg 33 , Arg 36 , and Phe 65 ), whereas additional, receptor-specific residues bind selectively to either the muscle (Lys 23 and Lys 49 ) or to the ␣7 (Ala 28 , Cys 26 -Cys 30 , and Lys 35 ) receptors. Of these, all but Cys 26 -Cys 30 and Lys 35 have their side chains accessible from the concave side of the flat, leaf-shaped toxin, indicating that both surfaces of ␣-cobratoxin appear to be involved in binding to the ␣7 receptor, in contrast to its interaction with the muscle ␣␤␥␦ receptor that involves only its concave surface (19,22,23). Of the 10 functionally important residues for ␣7 receptors, only Trp 25 , Ala 28 , Arg 33 , and Arg 36 are present in homologous positions in candoxin (Fig. 6C). These four residues have their side chains oriented toward the concave surface of candoxin, in agreement with their disposition in ␣-cobratoxin (Fig. 6D). Significantly, Arg 33 , reported to be the most crucial residue for the binding of long chain ␣-neurotoxins to ␣7 receptors (23) as well as for -bungarotoxins to bind to neuronal ␣3␤2 receptors (62), is present in candoxin (Arg 35 ). Interestingly, the sequence of WTX (24), which showed poor affinity for both receptors subsets, has only Lys 27 and Lys 50 , which are critical for the muscle ␣␤␥␦ receptor, and Arg 37 , which is involved in the recognition of both receptor types. However, although several (ϳ40%) functionally important residues for muscle (␣␤␥␦) and ␣7 receptors are present in homologous positions in candoxin, their precise roles in any interaction with these receptors will have to be determined by site-directed mutagenesis studies.
In conclusion, candoxin is a novel three-finger toxin that is a reversible antagonist of muscle (␣␤␥␦) but a poorly reversible antagonist of neuronal ␣7 nicotinic receptors. It is likely that candoxin, which lacks the helix-like conformation of the tip of the middle loop seen in long chain neurotoxins and hitherto considered essential for high affinity binding to ␣7 receptors (17), may have other functional determinants that account for its antagonism of ␣7 receptors in low nanomolar concentrations.