Discovery and characterization of a family of insecticidal neurotoxins with a rare vicinal disulfide bridge

Abstract

We have isolated a family of insect-selective neurotoxins from the venom of the Australian funnel-web spider that appear to be good candidates for biopesticide engineering. These peptides, which we have named the Janus-faced atracotoxins (J-ACTXs), each contain 36 or 37 residues, with four disulfide bridges, and they show no homology to any sequences in the protein/DNA databases. The three-dimensional structure of one of these toxins reveals an extremely rare vicinal disulfide bridge that we demonstrate to be critical for insecticidal activity. We propose that J-ACTX comprises an ancestral protein fold that we refer to as the disulfide-directed β-hairpin.

Main

Agricultural pesticide management is becoming increasingly complicated due to the evolution of insect resistance to classic chemical pesticides¹, growing community awareness of the environmental damage caused by many agrochemicals, and strict limits enforced by many countries on the level of pesticide residuals in imported crops and livestock. These concerns have stimulated the search for ‘environmentally friendly’ pest-control strategies.

One option, although not without potential problems such as accelerated insect resistance² and transgene escape³, is to engineer insect-specific toxins into plants. This is exemplified by the engineering of genes encoding insecticidal δ-endotoxins from the soil bacterium Bacillus thuringiensis into a variety of agriculturally important cultivars. A potentially more selective method is to use insect-specific viruses as vectors to deliver toxins to a restricted number of target species without harming beneficial insects and predators of the targeted pest. For example, most baculoviruses target only lepidopterans, an order that includes some of the most refractory agricultural pests, and this host specificity is not altered by expression of heterologous toxins^4,5,6,7.

Unfortunately, there are very few well-characterized insect-specific biopesticides that lend themselves to genomic approaches. We have thus embarked on a program that aims to discover and characterize insect-specific peptide toxins suitable for engineering into plants and viruses. Here we report on a family of three homologous insecticidal neurotoxins isolated from the venom of one of the world's most lethal arachnids⁸, the Blue Mountains funnel-web spider Hadronyche versuta. These peptides contain 36 or 37 residues and 4 disulfide bonds, including an extremely rare vicinal disulfide — a disulfide bond between adjacent cysteine residues — that we show is essential for insecticidal activity. The toxins can be folded nonenzymatically in vitro with 100% efficiency despite the presence of four disulfide bridges, enhancing the likelihood of successful heterologous expression of the toxins in plants and virus infected insects. The three-dimensional solution structure of one of these peptides reveals an ancestral protein fold that we refer to as the disulfide-directed β-hairpin (DDH).

Isolation of toxins

A typical reverse phase (rp)HPLC fractionation of crude venom from H. versuta is shown (Fig. 1a). The venom is highly complex, and all but the earliest eluting peaks (R_t< 9 min) are peptides of mass less than 10 kDa (ref. 11). An expansion of a portion of the chromatogram (inset, Fig. 1a) shows three peptides with similar retention times. These peptides proved to be highly insecticidal (Fig. 2) but were inactive on vertebrate smooth and skeletal nerve muscle preparations (see below).

Figure 1: Isolation and sequencing of J-ACTXs.

a, Reverse phase HPLC chromatogram of whole H. versuta venom; the inset is an amplified view illustrating the retention times of the insecticidal toxins J-ACTX-Hv1a, J-ACTX-Hv1b, and J-ACTX-Hv1c. b, Comparison of primary structures⁴⁹ of the J-ACTXs. Homologies are shown relative to J-ACTX-Hv1a; identities are boxed in yellow, and conservative substitutions are shown in red. The disulfide bonding pattern determined for J-ACTX-Hv1c from NMR and chemical analyses is shown below the sequence; note that the eight cysteine residues, and by inference the pattern of disulfide bridges, are strictly conserved in all three toxins. The secondary structure of J-ACTX-Hv1c, as determined in the current study, is shown above the sequences (β-strands in blue and β-turns in green, with the type of β-turn indicated by the numerals inside the boxes).

Figure 2: Dose-response curves from injection of J-ACTXs into house crickets.

Each point represents the mean of 2–3 independent experiments. The data were fit as described¹¹ to yield LD₅₀ values of 303 ± 42, 214 ± 16, 167 ± 10 pmol g⁻¹ for J-ACTX-Hv1a, J-ACTX-Hv1b, and J-ACTX-Hv1c, respectively.

Sequencing and mass spectral analysis revealed that the three peptides are highly homologous, each is 36 or 37 residues in length with eight conserved cysteine residues involved in four disulfide bridges (Fig. 1b). We have named these peptides the J-atracotoxins (J-ACTXs); the sequences, which show no homology to any in the protein and DNA sequence databases, have been deposited in the SWISS-PROT data bank (accession numbers P82226–P82228). The LD₅₀ values for these toxins in house crickets (167–303 pmol g⁻¹; Fig. 2) are similar to those previously obtained for the ω-atracotoxins¹¹.

In vitro folding of J-ACTXs

The J-ACTXs are minor components of funnel-web venom and hence sufficient amounts of native material could not be purified for detailed structural and functional characterization. Consequently, we attempted to fold synthetic J-ACTX-Hv1c (the most potent of the J-ACTXs) in vitro. Most insecticidal toxins being considered for genomic approaches to pest management have a high disulfide bridge content, such as insect-specific scorpion α-toxins^4,12, which, like J-ACTX, contain four intramolecular disulfide bonds and hence 105 possible disulfide isomers. Proper oxidation/folding of these toxins in transgenic plants or virus infected insects is critical as their effectiveness could be markedly impaired if folding is inefficient. In vitro folding of these toxins is likely to give some indication of their intrinsic, sequence-dependent, nonenzymatic folding potential.

The time course of oxidation/folding of synthetic J-ACTX-Hv1c in a simple glutathione redox buffer devoid of protein folding machinery such as molecular chaperones or disulfide isomerases is shown in Fig. 3a. Remarkably, the folding reaction is almost complete within two hours and the final product, which co-elutes with native J-ACTX-Hv1c, is obtained with a 100% yield. This product was found to be as potent as native J-ACTX-Hv1c in insect bioassays, and its one-dimensional ¹H NMR spectrum was identical to that of native toxin (data not shown), indicating that the native fold had been obtained. Interestingly, the folding reaction seems to occur via a discrete pathway as indicated by the three distinct peaks evident between the reduced and oxidized forms in the HPLC chromatogram, which may correspond to intermediates with one, two and three disulfides.

Figure 3: In vitro folding and disulfide bridge pattern of J-ACTX-Hv1c.

a, A series of rpHPLC chromatograms illustrating the in vitro folding of J-ACTX-Hv1c. The reduced peptide was dissolved in a redox buffer that promotes disulfide shuffling, then aliquots of the reaction mixture were examined over a period of 48 h. The final product, which co-elutes with native J-ACTX-Hv1c purified from the venom of H. versuta, is obtained with 100% yield. b, Oxidized J-ACTX-Hv1c obtained from the in vitro folding reaction was partially reduced at low pH then the reduced cysteines were carboxamidomethylated by reaction with iodoacetamide. The two chromatograms illustrate rpHPLC fractionation of the carboxamidomethylated species that were obtained following reduction at the indicated temperatures and time intervals. These carboxamidomethylated species were then fully reduced and the remaining cysteines were pyridethylated before the peptides were sequenced. The sequences corresponding to two of the species are indicated, with green circles and orange boxes indicating carboxamidomethylated and pyridethylated cysteines respectively. In combination, these sequences confirm the presence of Cys 10–Cys 22 and Cys 13–Cys 14 disulfide bonds in J-ACTX-Hv1c.

Three-dimensional structure of J-ACTX-Hv1c

The solution structure of the folded, synthetic J-ACTX-Hv1c was determined using standard homonuclear NMR methods¹³. Final structure calculations employed ∼13 restraints per residue (Table 1). An ensemble of 20 structures with the lowest residual restraint violations was used to represent the solution structure of the toxin (Fig. 4, Table 1). The structure is very precise, with backbone and heavy atom root mean square (r.m.s.) deviation values of 0.18 ± 0.06 Å and 0.57 ± 0.09 Å, respectively, for the well-defined region (residues 3–34). According to PROCHECK¹⁴, 76% of the non-Pro/Gly residues in the structured region lie in the most favored sector of the Ramachandran plot, with the remaining 24% located in the additionally allowed regions.

Table 1 Structural statistics for the family of 20 J-ACTX-Hv1c structures

Figure 4: Solution structure of J-ACTX-Hv1c.

a, Stereo view of the ensemble of 20 J-ACTX-Hv1c structures superimposed for best fit over the backbone atoms of residues 3–34 of the mean coordinate structure. The side chain heavy atoms of the disulfide bonds (Cys 3–Cys 17, Cys 10–Cys 22, Cys 13–Cys 14, and Cys 16–Cys 32) are shown in red, and the backbone is colored green (β-turns), gold (β-strands), and blue (nonregular structure). The three central disulfide bridges form an inhibitory polypeptide cystine knot motif^15,16. b, Same view as in (a) with the side chain heavy atoms of all the hydrophobic residues in J-ACTX-Hv1c shown in pink. Note that these residues cluster on the face of the molecule containing the vicinal disulfide bridge (the left hand surface in this view). c, Schematic of J-ACTX-Hv1c illustrating the location of the β-strands (gold), β-turns (green), and disulfide bridges (red). The molecule can be conveniently pictured as a series of loops (numbered 1–4 from the N-terminus to the C-terminus) bounded by half-cystine residues.

The structure of J-ACTX-Hv1c consists of a disulfide rich globular core comprising residues 3–19, with residues 20–34 forming a β-hairpin that projects from this region. Residues 1–2 and 35–37 at the N-terminus and C-terminus, respectively, are disordered in solution. The toxin has a high content of regular secondary structure (∼80% if the disordered N- and C-termini are excluded) as most of the polypeptide chain outside of the two β-strands (yellow in Fig. 4) is involved in various types of β-turns (green). The three central disulfide bonds form an inhibitory cystine knot (ICK) motif^15,16 in which the Cys 16–Cys 32 disulfide passes through a 14-residue ring formed by the Cys 10–Cys 22 and Cys 3–Cys 17 disulfide bridges and the intervening sections of polypeptide backbone.

J-ACTX-Hv1c can be conveniently considered as comprising four loops (numbered 1–4 from the N-terminus to the C-terminus; Fig. 4c), each bounded by half-cystine residues. The hydrophobic core of the globular region is composed of the Cys 10–Cys 22 and Cys 16–Cys 32 disulfide bridges as well as the side chains of Thr 4 and Thr 20 (Fig. 4b). We previously noted that the hydrophobic core of cystine knot toxins is formed largely by the two C-terminal disulfides and a single hydrophobic residue protruding from loop 1 or 3 (Ile 5 in the case of ω-ACTX-Hv1a)¹⁷. The hydrophobic residues of J-ACTX-Hv1c are conspicuously clustered on one face of the molecule (Fig. 4b).

The vicinal disulfide bridge

The most remarkable feature of the structure of J-ACTX-Hv1c is the vicinal disulfide bridge connecting Cys 13 to Cys 14. While the disulfide configuration appeared unambiguous from the NMR data, the rarity of vicinal disulfide bonds in proteins prompted us to seek chemical confirmation. We used partial reduction at low pH followed by alkylation^17,18 to trap partly reduced intermediates that would reveal the disulfide bond pattern of J-ACTX-Hv1c. In this method, partially reduced intermediates are rapidly alkylated with iodoacetamide, then the peptide is fully reduced and the remaining cysteine residues are pyridethylated. The peptides are then sequenced and the position of the disulfide bonds can be inferred from the location of the pairs of carboxamidomethylated and pyridethylated cysteines.

Two HPLC chromatograms obtained from reduction of J-ACTX-Hv1c at different temperatures are shown in Fig. 3b. Sequencing of one of the intermediates from the reaction at 37 °C yielded a peptide that was carboxamidomethylated at Cys 10, 13, 14, and 22 and pyridethylated at Cys 3, 16, 17, and 32, while one of the intermediates isolated from the reaction at 15 °C proved to be carboxamidomethylated at only Cys 10 and 22 (Fig. 4b). In combination, these intermediates provide evidence for the Cys 10–Cys 22 and Cys 13–Cys 14 disulfide bonds, thus providing chemical confirmation of the vicinal disulfide bridge.

Configuration of the vicinal disulfide bridge

Vicinal disulfide bridges in proteins are extremely rare; apart from J-ACTX, the only examples thus far reported are found in methanol dehydrogenase (MDH)¹⁹ and the α-subunit of the acetylcholine receptor (αAChR)²⁰. Theoretical calculations^21,22 predicted that the eight-membered ring formed as a result of disulfide formation between adjacent cysteines would be most stable if the intervening peptide bond was in a slightly nonplanar cis-like configuration; studies on model peptides²³, as well as a low resolution structure of MDH¹⁹, appeared to confirm this view.

However, the structure of J-ACTX-Hv1c, as well as more recent high resolution structures of MDH, present a striking disparity from these theoretical and model compound analyses. A comparison of the vicinal disulfide bridge of J-ACTX-Hv1c with that from the 1.9 Å resolution crystal structure of MDH from Methylophilus W3A1 (ref. 24) is shown in Fig. 5; the atoms constituting the eight-membered ring overlay remarkably well with an r.m.s. deviation of 0.18 Å. The individual torsion angles of the disulfide ring are listed in the lower panel of Fig. 5 for J-ACTX-Hv1c, MDH²⁴, and a disulfide bridged Cys–Cys dipeptide (cyclo-Cys-Cys)²³. The Cys–Cys peptide bonds in MDH and J-ACTX-Hv1c are nonplanar (ω = −147° and −135°, respectively) but, in contrast to cyclo-Cys-Cys, they are much closer to trans than cis. The disulfide chirality is left handed for both MDH and J-ACTX-Hv1c (χ₃ = −106°), again in striking contrast to cyclo-Cys-Cys where the chirality is reversed (χ₃ = 94°).

Figure 5: Examining the disulfides.

a, Stereo view of the Cys 13–Cys 14 disulfide bridge of J-ACTX-Hv1c (backbone and cystine side chains in light and dark green, respectively) overlaid for best fit with the Cys 103–Cys104 vicinal disulfide bridge from the 1.9 Å resolution structure (PDB accession code 1B2N) of Methylophilus W3A1 MDH²⁴ (backbone and cystine side chains in pink and orange, respectively). The heavy atoms of the pair of linked cystine residues overlay with an r.m.s. deviation of 0.18 Å. b, The individual torsion angles of the eight-membered disulfide ring for a disulfide bridged Cys–Cys dipeptide (Cys-Cys)²³, MDH²⁴, the best J-ACTX-Hv1c structure (b), and the average of the ensemble of 20 J-ACTX-Hv1c structures (a). The final column lists the Cα–Cα distances for these disulfide bridges.

The 1.94 Å resolution structure of Methylobacterium extorquens MDH²⁵ reveals a highly nonplanar peptide bond for the vicinal disulfide bridge with a deviation of −35° from the trans configuration (that is, ω = 145°), and a recent NMR study of α-conotoxin GI revealed that the peptide bond was close to trans in a mutant with a non-native vicinal disulfide²⁶. Thus, while only a few examples are currently available from which to formulate general conclusions, the emerging consensus appears to be that the strain imparted by formation of a vicinal disulfide bridge in native proteins is alleviated by the intervening peptide bond assuming a distorted trans configuration.

The vicinal disulfide is functionally important

The three nonvicinal disulfide bridges in J-ACTX-Hv1c connect residues that are distal in the sequence; these disulfides essentially determine the tertiary fold of cystine knot toxins²⁷. The Cys 13–Cys 14 vicinal disulfide, on the other hand, is unlikely to play an important architectural role. Instead, the vicinal disulfide bonds in MDH and αAChR have critical functional roles: MDH activity is abolished when the vicinal disulfide at the enzyme active site is reduced¹⁹ and the vicinal disulfide bridge in αAChR is essential for acetylcholine binding²⁸.

In order to test the functional significance of the vicinal disulfide in J-ACTX-Hv1c, we constructed a double mutant in which Cys 13 and Cys 14 were both replaced with Ser. The ¹H NMR chemical shifts of the mutant toxin were almost coincident with those of native J-ACTX-Hv1c (with the exception of the mutated residues), indicating that the tertiary fold was unaffected by the point mutations. As additional proof we calculated a medium resolution backbone structure of the mutant toxin using 272 distance and 23 φ-angle restraints obtained from a single NOESY experiment. The mutant and native toxin structures overlay with a backbone r.m.s. deviation of 0.42 Å, and the overall conformation of loop 2 is largely unaffected by the two mutations in this region. However, the double mutant was completely inactive when injected into crickets at doses up to 60 times the LD₅₀ of the native toxin (Fig. 2), demonstrating that the vicinal disulfide is critical for insecticidal activity. Thus, vicinal disulfide bridges play key functional, rather than architectural, roles in all three proteins in which they have been discovered thus far.

Is the DDH motif an ancestral protein fold?

A search of the Protein Data Bank²⁹ revealed a number of structural homologs of J-ACTX-Hv1c, most of which are members of the ICK family of toxic polypeptides^15,16. With one exception, the three best matches are with toxins that modulate voltage gated sodium channels (Fig. 6a): δ-atracotoxin^27,30 from the Australian funnel-web spider; μ-agatoxin I (ref. 31) from the unrelated American funnel-web spider Agelenopsis aperta; and conotoxin GS³² from the aquatic cone snail Conus geographus.

Figure 6: Structural homologs of J-ACTX-Hv1c and examples of DDH motifs.

a, Overlay of J-ACTX-Hv1c (pink) on the structures of the sodium channel toxins δ-ACTX-Hv1a (green; PDB accession code 1VTX)²⁷, μ-agatoxin-I (gold; PDB accession code 1EIT)³¹, and conotoxin GS (blue; PDB accession code 1AG7)³². The core ICK region (that is, the two β-strands, three disulfide bridges, and loops 1 and 3) of J-ACTX-Hv1c can be overlaid on the backbone and cystine side chain heavy atoms of the corresponding regions of conotoxin-GS, μ-agatoxin-I, and δ-ACTX-Hv1a with r.m.s. deviation values of 1.20 Å, 1.04 Å, and 1.09 Å, respectively. The folds are highly homologous with hypervariability evident only in loop 2. J-ACTX-Hv1c is rotated 180° around the long axis of the β-strands relative to the views in Fig. 4. b, Overlay of the structures of J-ACTX-Hv1c (pink) and the cellulose binding domain (green) of fungal cellobiohydrolase I (PDB accession code 2CBH)³³. The orientation of J-ACTX-Hv1c is similar to that in Fig. 4. The backbone of residues 4–10, 15–25, and 29–32 plus the heavy atoms of Cys 10, 16, 22, and 32 of J-ACTX-Hv1c can be overlaid on the backbone of residues 2–8, 18–28, and 32–35 plus the heavy atoms of Cys 8, 19, 25, and 35 of CBD-CBHI with an r.m.s. deviation of 0.67 Å. The only major divergence between these two structures is the greater elaboration of loop 2 in CBD-CBHI. c, Schematic of the disulfide directed β-hairpin fold showing the position of β-strands, disulfide bridges, and various loops. Note that loop 3 generally conforms to the consensus X₂-G-X₂. d, Sequences of selected eukaryotic DDH domains. Domains are grouped into those that occur in isolation (I), in duplicated form (D), or as fusions with other protein domains (F). The structures of all these domains have been determined except Dickkopf-1 (Dkk-1), whose inclusion in this grouping is speculative. Note the marked conservation of glycine in the central position of loop 3, and the common occurrence of a hydrophobe in the subsequent position (highlighted in red). ‘Nterm’ and ‘Cterm’ indicate the N-terminal and C-terminal DDH folds in proteins where the domain has been duplicated. Gurmarin is a sweet taste suppressor from the plant Gymnema sylvestre³⁹ and the cellobiohydrolase I sequence is from T. reesei.

Curiously, however, the closest structural homolog of J-ACTX-Hv1c is the cellulose binding domain of cellobiohydrolase I (CBD-CBHI) from the fungus Trichoderma reesei (Fig. 6b), which is missing the N-terminal disulfide bridge of the ICK motif. We have shown that the compact hydrophobic core of the ICK motif consists largely of the two disulfide bridges that emanate from the two β-strands that characterize the ICK fold¹⁷. The N-terminal disulfide bridge contributes very little to the hydrophobic core and, as indicated by the structure of CBD-CBHI³³, is not essential to formation of the basic ICK fold. Indeed, it has been demonstrated that the tertiary structure and thermal stability of the ICK-containing trypsin inhibitor EETI II is largely unperturbed by removal of the N-terminal disulfide bridge^9,34.

We propose that the ICK fold is actually a minor elaboration of a simpler ancestral fold that we refer to as the disulfide-directed β-hairpin (DDH) fold, of which CBD-CBHI is the archetypal family member. The DDH fold is shown schematically in Fig. 6c and its amino acid consensus sequence can be written as CX_5–19CX₂[G or P]X₂CX_6–19C, where X is any amino acid (Fig. 6d). The DDH fold differs from the ICK fold in that there are only two mandatory disulfide bridges (the two that form the bulk of the hydrophobic core), so that loop 1 as shown in Fig. 4c is no longer necessarily bounded by an N-terminal cysteine residue, and loop 3 is generally five residues in length with a central Gly or Pro to ensure a tight turn prior to the first β-stand. The residue following the Gly/Pro in loop 3 is generally hydrophobic. This residue along with the two buried disulfide bridges constitute the mini-hydrophobic core of this domain.

Several vertebrate proteins appear to have arisen from simple duplication of an ancestral DDH gene followed by minor loop elaborations. The ICK motif, in contrast, has not been found in vertebrates. An overlay of two molecules of J-ACTX-Hv1c on the structure of a single molecule of mamba intestinal toxin I (MIT1) from the venom of the snake Dendroaspis polylepis polylepis³⁵ is shown in Fig. 6a. It is clear that one can obtain MIT1 by simple head-to-tail duplication of a J-ACTX-like gene followed by elaboration of loop 2 (particularly in the C-terminal DDH domain). Two molecules of J-ACTX-Hv1c can also be overlaid equally well on colipase, a molecule that is widely distributed in vertebrates and has been noted to be a structural homolog of MIT1³⁵. However, as shown in the overlay of J-ACTX-Hv1c on the C-terminal DDH domain of porcine colipase in Fig. 7b, the structural similarities can sometimes be obscured by significant loop elaborations such as the helical insertion in loop 4 of colipase.

Figure 7: Comparison of J-ACTX-Hv1c with vertebrate DDH domains.

a, Overlay of two molecules of J-ACTX-Hv1c (pink and gold) on the structure of a single molecule of MIT1 (PDB accession code 1MIT)³⁵ shown in green. The backbone of residues 3–10, 14–25, and 29–34 plus the heavy atoms of Cys 10, 16, 22, and 32 of J-ACTX-Hv1c can be overlaid on the backbone of residues 34–41, 59–70, and 74–79 plus the heavy atoms of Cys 41, 61, 67, 77 of MIT1 with an r.m.s. deviation of 1.15 Å. Similarly, the backbone of residues 9–10, 15–24, and 27–34 plus the heavy atoms of Cys 10, 16, 22, and 32 of J-ACTX-Hv1c can be overlaid on the backbone of residues 6–7, 12–21, and 26–33 plus the heavy atoms of Cys 7, 13, 19, and 31 of MIT1 with an r.m.s. deviation of 1.17 Å. MIT1 comprises two minimal DDH folds (as exemplified by the structure of J-ACTX-Hv1c) held together largely by an elaboration of loop 2 in the C-terminal DDH domain, which serves to tether the two domains. b, Stereo view of an overlay of J-ACTX-Hv1c on the C-terminal DDH domain of porcine colipase (PDB accession code 1LPB)⁵⁰. The backbone of residues 3–10, 16–23, and 30–34 plus the heavy atoms of Cys 10, 16, 22, and 32 of J-ACTX-Hv1c can be overlaid on the backbone of residues 42–49, 63–70, and 85–89 plus the heavy atoms of Cys 49, 63, 69, and 87 of porcine colipase with an r.m.s. deviation of 0.88 Å. A similar overlay can be made of J-ACTX-Hv1c on the corresponding atoms of the N-terminal DDH domain of porcine colipase (r.m.s. deviation of 1.34 Å).

The DDH fold differs from the previously proposed T-knot scaffold³⁶ in that the sequence and topology are more rigidly defined for loop 3, and there is no requirement for loops 2 and 4 to have any specific orientation with respect to the β-hairpin. The DDH fold is similar to the recently proposed cystine-stabilized β-sheet (CSB) motif⁹. The CSB motif, however, is more rigidly defined as a triple-stranded β-sheet having the consensus sequence CX_4–6CX_2–6CX_5–10C, which excludes many of the DDH containing proteins in Fig. 6d. Indeed, to date, no native proteins have been found that contain the elementary CSB motif without additional disulfides⁹.

Evolution of the DDH fold

Surprisingly, an exhaustive search of the protein and DNA sequence data bases, including numerous partially sequenced bacterial genomes, failed to identify any DDH motifs in the Archaea or Eubacteria. The DDH fold is, however, present in a wide variety of eukaryotes, including fungi, red algae (rhodophytes), arachnids, aquatic snails, plants, and vertebrates. This scenario tentatively suggests that this cystine framework evolved in an ancestral eukaryote prior to the divergence of plants, animals, and fungi, and that the marked sequence variability simply reflects the tolerance of this fold to sequence changes.

We have detected numerous ancestral duplications of this motif, as well as functional associations with other protein domains. The fungal CBD-CBHI domain is part of a multi-domain structure, and the homologous domain in the putative polysaccharide binding protein of the rhodophyte Porphyra purpurea occurs as four consecutive repeats without significant loop variability (see Pfam alignment CBD_1). Gene duplication followed by the same type of loop elaborations as observed in vertebrates can be seen in ACTX-Hvf17 from the Australian funnel-web spider¹⁰, and this duplicated motif also appears in the vertebrate embryonic head inducer Dickkopf-1 (ref. 37) where it appears to represent the N-terminal domain of a multi-domain protein¹⁰. Thus, the DDH fold appears to be widely distributed in eukaryotes, and can occur as a single domain (for example, ICK polypeptides), in duplicated form (for example, MIT1 and colipase), or as a fusion with other protein domains (for example, fungal cellobiohydrolases and insulin-like growth factor (IGF) binding proteins).

What is the physiological target of J-ACTX?

The J-ACTXs seem to be highly insect specific; at concentrations as high as 1 μM, J-ACTX-Hv1c had no effect on electrically stimulated contractions of vertebrate smooth (rat vas deferens) or skeletal (chicken biventer cervicis) muscle (data not shown), nor did it cause any adverse effects when injected into newborn mice at doses up to 3.14 μg g⁻¹, which is five-fold higher than the LD₅₀ in A. domesticus. However, direct application of J-ACTX-Hv1c (0.5–1.0 nmol) to the metathoracic ganglion of the cockroach Periplaneta americana caused spontaneous, uncoordinated movement of all legs within 60 s, which developed into fasciculations of the limbs and posterior sensory organs (cerci) within 1–3 min. Thus, J-ACTX-Hv1c appears to be an excitatory neurotoxin whereas, in marked contrast, ω-ACTX-Hv1a can be considered a depressant neurotoxin since it blocks reflex movements of the hind legs in this assay¹⁷.

Scorpion excitatory neurotoxins act at insect voltage gated sodium channels⁵. However, despite the similarity in tertiary fold between J-ACTX and the sodium channel modulators μ-agatoxin-I, δ-ACTX-Hv1, and conotoxin GS (Fig. 6a), we found that J-ACTX had no effect on whole cell potassium, sodium, or calcium currents in isolated bee brain neurons (data not shown). We also addressed the intriguing possibility that J-ACTX might mimic the α-subunit of the AChR, which also contains a vicinal disulfide bridge that forms part of the ACh binding site; AChR activity could be disrupted by J-ACTX perturbing α-subunit interactions and/or ACh binding. We found that J-ACTX-Hv1c did not affect steady state acetylcholine currents in bee brain neurons; however, it remains possible that J-ACTX antagonizes muscle AChRs or AChR subtypes present in other invertebrate neurons.

The putative bioactive surface of J-ACTX

J-ACTX possesses two strikingly dissimilar faces (Fig. 8). One face presents an almost contiguous charged surface, while the opposing face containing the vicinal disulfide is devoid of ionizable side chains (Figs 4b, 8). Thus, in the absence of a molecular target, we have tentatively named these toxins the Janus-faced atracotoxins or J-atracotoxins.

Figure 8: Electrostatic surface potential of J-ACTX-Hv1c with positive and negative charges shown in blue and red, respectively.

The orientation is as for Fig. 4. Note the striking asymmetry in charge distribution; the surface of the molecule containing the vicinal disulfide bridge is devoid of ionizable residues, while the opposite face presents an almost contiguous charged surface. Note the hydrophobic cleft and nearby cavity (Cav), which exposes the side chain hydroxyl of Thr 4.

The critical functional role of the vicinal disulfide indicates that this feature is most likely present at the toxin's bioactive surface. The region surrounding the vicinal disulfide displays several features that suggest it might represent the bioactive surface of the toxin (Fig. 8). First, this region contains several exposed hydrophobic residues that are conserved in all three J-ACTXs (Ala/Thr 1, Ile 2, Ala 12, Cys 13, Cys 14, and Pro 15). Second, the side chains of Ala 12–Pro 15 form a wall that borders a hydrophobic cleft overhung by the methyl groups of Val 29. At one end of this cleft is a large cavity (Fig. 8) that might accommodate chemical groups of the target molecule.

What is the role of the vicinal disulfide bridge? One possibility is that the vicinal disulfide, while clearly not important in determining the overall fold of J-ACTX, might play a local architectural role, such as facilitating configuration of the hydrophobic cleft. However, the structure of loop 2 is largely unaltered in the Ser–Ser mutant, which argues against this hypothesis. It is possible that the redox potential of the vicinal disulfide, because of its strained cyclic geometry, might be sufficiently altered from that of typical disulfides to facilitate covalent reaction of these cysteines with sites on the target molecule. If this were the case, we might expect this disulfide to be particularly susceptible to reduction. However, the Cys 10–Cys 22 disulfide was the first to be reduced in partial reduction experiments (Fig. 3b). The vicinal disulfide bridge does not, therefore, appear to be unusually reactive. Thus, we speculate that the vicinal disulfide bridge is directly involved in interactions with the target molecule, just as the vicinal disulfide bridge in the AChR appears to be directly involved in acetylcholine binding.

Methods

Purification of toxins

Lyophilized crude venom was loaded onto a Vydac C18 analytical rpHPLC column and eluted at flow rate of 1 ml min⁻¹ using a gradient of 5–25% buffer B (0.1% v/v trifluoroacetic acid (TFA) in acetonitrile) over 22 min, followed by a gradient of 25–50% buffer B over 48 min. Buffer A was 0.1% v/v TFA in water. Fractions were collected manually, and individual components were further purified using shallower acetonitrile gradients. Once purified to > 98% homogeneity (as assessed by rpHPLC), peptides were lyophilized and stored at −20 °C until further use. Peptides were pyridethylated as described¹¹ prior to sequencing on an Applied Biosystems 473 protein sequencer.

Insect bioassays

Peptides were considered insect active if they caused death or prolonged paralysis (>6 h) when injected into mealworms (Tenebrio molitor) and/or house crickets (Acheta domesticus Linnaeus)¹¹. For quantitative analysis of insecticidal activity, the LD₅₀ in A. domesticus was determined as described¹¹.

The effect of J-ACTX-Hv1c on neural transmission was examined using an assay as described¹⁷. Briefly, the abdominal and thoracic ganglia were exposed by removal of the dorsal surface and viscera but not the cerci or legs of immobilized P. americana. The toxin was applied directly to the metathoracic ganglion, which was isolated by means of a polyethylene well sealed into place with agar. In this bioassay, stimulation of the insect's posterior sensory organs (cerci) in the absence of toxin elicits reflex movement of the cockroach legs.

Vertebrate bioassays

Peptide toxins were assayed for their effect in vertebrate smooth (vas deferens) and skeletal (biventer cervicis) nerve muscle tissue as described¹⁰. Contractions of each tissue were recorded in the absence of additives or following injection of peptides directly into the bath buffer. δ-ACTX-Hv1 (100 nM), a potent modulator of voltage gated sodium channels³⁸, was used as a positive control.

Vertebrate toxicity assay

Vertebrate toxicity of J-ACTX-Hv1c was determined by subcutaneous injection of toxin in 0.1 ml saline into young BALB/c mice (2.9 ± 0.1 g). Acute toxicity tests were conducted with a minimum number of animals and toxicity was monitored over a period of 48 h.

Bee brain dissociation and electrophysiology.

Neurons were dissociated from the brains of adult European honeybees (Apis mellifera) using a procedure that will be described in detail elsewhere. Briefly, isolated bee brains were incubated for 2 min at room temperature in buffer containing 20 units ml⁻¹ papain. Digestion was terminated with a solution containing 1 mg ml⁻¹ ovomucoid papain/trypsin inhibitor (Type II-O, Sigma) and 1 mg ml⁻¹ bovine serum albumin. Neurons were isolated by gentle trituration though a series of decreasing bore silanized pasteur pipets with fire-polished tips, and then plated on plastic culture dishes coated with poly-d-lysine.

Whole cell voltage clamp recordings were made of bee brain nicotinic AChR and Ca²⁺, Na⁺, and K⁺ channel currents, characterization of which will be reported elsewhere. Neurons were voltage clamped at −90 mV and currents evoked by stepping the membrane potential to −60 mV. Toxin effects on Ca²⁺ and Na⁺ currents were tested at the potential with the largest inward current, usually −10 or 0 mV. Toxin effects on K⁺ channels were determined over a range of membrane potentials (−60 to +60 mV). Nicotinic AChR currents (induced with 30 mM ACh) were obtained at −90 mV. Data were collected and analyzed off line with the PCLAMP and Axotape suite of programs (Axon Instruments).

Folding and purification of synthetic toxins

Synthetic J-ACTX-Hv1c and a Cys 13–Cys 14→Ser-Ser mutant of this toxin were purchased from Auspep. The reduced peptides were purified to homogeneity using rpHPLC, then oxidized/folded in a glutathione redox buffer that promotes disulfide oxidation and shuffling³⁹. At various times after initiation of the folding reaction, aliquots were removed, quenched with HCl, and analyzed using rpHPLC in order to monitor progress of the folding reaction. After 48 h, the reaction mixture was quenched and dialyzed against H₂O to remove folding buffer components. The lyophilized dialyzate was then dissolved in H₂O and the fully oxidized J-ACTX-Hv1c was purified using rpHPLC.

Chemical confirmation of the vicinal disulfide bond

The disulfide bonding pattern of J-ACTX-Hv1c was determined as described^17,18. Briefly, fully oxidized J-ACTX-Hv1c was partially reduced by incubation with a 1:40 molar ratio of Tris-(2-carboxyethyl)-phosphine at pH 3.0 for 15–60 min at either 4 °C, 15 °C, or 37 °C. The reduced cysteines were then carboxamidomethylated by rapid addition of a supersaturated solution of iodoacetamide. The partially reduced/alkylated species were then separated using rpHPLC and their molecular weights determined using mass spectrometry to ascertain the number of cysteines that had been carboxamidomethylated. The peptides were then fully reduced and the unmodified cysteines were pyridethylated. Peptides containing potentially informative combinations of modified cysteine residues were then sequenced on an Applied Biosystems 473 protein sequencer to determine the positions of the carboxamidomethylated and pyridethylated cysteines, from which the positions of disulfide bonds could be inferred.

NMR spectroscopy

NMR samples were prepared by dissolving 1.5 mg of J-ACTX-Hv1c or 2.9 mg of the Ser 13-Ser 14 mutant in 250 μl H₂O in a susceptibility-matched microcell (Shigemi) and adjusting the pH to 4.95. NMR spectra were recorded at 25 °C and 600 MHz using a 5 mm ¹H probe on a Bruker DRX-600 spectrometer. The following homonuclear 2D NMR spectra were recorded for J-ACTX-Hv1c: ECOSY, TOCSY with an MLEV isotropic mixing period of 70 ms, and NOESY with mixing times of 50 and 250 ms. A single NOESY spectrum with a mixing time of 300 ms was recorded for the Ser 13-Ser 14 mutant using a triple resonance inverse probe on a Varian INOVA 600 spectrometer.

Spectra were processed using XWINNMR (Bruker) or Felix97 (Molecular Simulations) and chemical shift assignments were made using XEASY⁴⁰. Slowly exchanging amides were identified by reconstituting lyophilized J-ACTX-Hv1c in 250 μl 99.96% D₂O (Sigma) and immediately recording a time course of 1D spectra for ∼60 min, followed by a time course of eight TOCSY spectra over 16 h.

Structure calculations

NOESY crosspeaks were integrated in XEASY and converted to distance restraints (with pseudoatoms where appropriate) using CALIBA⁴¹. This yielded 403 nonredundant distance restraints. Twenty dihedral angle restraints²⁷ were derived from ³J_HNHα coupling constants measured from either 1D NMR spectra or from inverse Fourier transforms of in-phase NOESY multiplets⁴². Seven additional φ restraints of −100 ± 80° were applied for residues for which the intraresidue Hα-HN NOE was clearly weaker than that between HN and the Hα of the preceding residue⁴³. The intense intraresidue Hα-HN NOE for Asp 7, combined with a ³J_HNHα value of ∼7 Hz, allowed its φ angle to be restrained to 50 ± 40° (refs 31,44).

Stereospecific assignment of β-methylene protons and χ₁ restraints were derived for 14 residues using Hα–Hβ coupling constants measured from ECOSY spectra in combination with Hα–Hβ and HN–Hβ NOE intensities⁴⁵ measured from the 50 ms NOESY experiment. The Hβ protons of Pro 9 and Pro 18 were stereospecifically assigned on the basis of NOE intensities⁴⁶; this was not possible for Pro 15 and Pro 37 as the β-protons were magnetically equivalent. All four X-Pro peptide bonds were clearly identified as trans on the basis of characteristic NOEs¹³. Thirteen slowly exchanging amide protons were unambiguously assigned as hydrogen bond acceptors on the basis of preliminary structure calculations. Corresponding hydrogen bond (i-j) restraints of 1.7–2.2 Å and 2.7–3.2 Å were employed for the HN_i–O_j and N_i–O_j distances, respectively. The disulfide bonding pattern was determined unequivocally from preliminary structure calculations and distance restraints¹⁷ were introduced for the four disulfide bridges in subsequent calculations.

The torsion angle dynamics program DYANA⁴¹ was used to calculate 5,000 structures from random starting conformations. The best 100 structures (selected on the basis of final penalty function values) were then refined in X-PLOR⁴⁷ as described²⁷. The network of experimental restraints was not compatible with either a cis (ω = 0°) or trans (ω = 180°) conformation for the Cys 13–Cys 14 peptide bond. Thus, this ω angle was not restrained in the structure calculations, thereby enabling its value to be determined solely by the network of experimental restraints. The 20 lowest energy X-PLOR structures were used to represent the solution structure of J-ACTX-Hv1c. Molecular graphics analyses were performed using MOLMOL⁴⁸.

Coordinates

Coordinates and experimental restraints for the ensemble of 20 J-ACTX-Hv1c structures have been deposited in the Protein Data Bank (PDB accession code 1DL0), and ¹H chemical shifts have been deposited in BioMagResBank (accession number 4685).