Adaptive Evolution of Scorpion Sodium Channel Toxins
- Cite this article as:
- Zhu, S., Bosmans, F. & Tytgat, J. J Mol Evol (2004) 58: 145. doi:10.1007/s00239-003-2534-2
- 256 Views
Gene duplication followed by positive Darwinian selection is an important evolutionary event at the molecular level, by which a gene can gain new functions. Such an event might have occurred in the evolution of scorpion sodium channel toxin genes (α- and β-groups). To test this hypothesis, a robust statistical method from Yang and co-workers based on the estimation of the nonsynonymous-to-synonymous rate ratio (ω = dN/dS) was performed. The results provide clear statistical evidence for adaptive molecular evolution of scorpion α- and β-toxin genes. A good match between the positively selected sites (evolutionary epitopes) and the putative bioactive surface (functional epitopes) indicates that these sites are most likely involved in functional recognition of sodium channels. Our results also shed light on the importance of the B-loop in the functional diversification of scorpion α- and β-toxins.
KeywordsGene duplicationPositive Darwinian selectionLikelihood ratio testScorpion toxinSodium channelEvolutionary epitopes
Scorpion sodium channel toxins are a class of small peptides composed of about 61–76 residues with four disulfide-bridges (Possani et al. 1999). All the members in this class share a similar gene organization and three-dimensional structure and therefore have probably evolved from a common ancestor (Froy et al. 1999; He et al. 1999). Based on their different pharmacological features, these polypeptides are divided into two distinct classes, called α- and β-toxins (Gurevitz et al. 1998; Cestèle and Catterall 2000). The α-toxins mainly cause a slowing of the inactivation process of sodium currents and a prolongation of the action potential by binding to receptor site 3 of the voltage-gated sodium channel (VGSC) (Catterall 1995; Catterall 2000; Denac et al. 2000; Gordon and Gurevitz 2003). This class of toxins can be further divided into three subgroups: the classical α-group, which is highly active in the mammalian brain (e.g. AaHII); the insect α-toxins (e.g., LqhαIT); and the α-like group, which is active in both mammalian and insect central nervous system (CNS) (e.g., BmKM1) (Gordon and Gurevitz 2003). The β-toxins cause the VGSC to shift the voltage dependence of activation to more negative membrane potentials and cause a reduction of peak current amplitude by binding to receptor site 4 (Cestèle et al. 2001). A voltage-sensor trapping mechanism was proposed to account for the effects of these toxins on channel gating (Cestèle et al. 2001). Scorpion β-toxins also display similar features to distinguish between mammalian and insect sodium channels (Possani et al. 1999).
Supported by these facts, it appears that scorpions have used a common scaffold to adapt to diverse pharmacological targets, which should be beneficial for catching their prey and maintaining their defence against predators (Froy et al. 1999).
However, what drives their divergence in the course of evolution? An emerging trend in the study of the evolution of multigene families is the observation of rapid, adaptive evolution driven by positive selection following gene duplication (Zhang et al. 1998; Yang et al. 2002). We hypothesize that it is a similar case for scorpion sodium channel toxin genes. To test our hypothesis, a new statistical method developed by Yang and co-workers was used. This method analyses variable selective pressures among sites by estimating the nonsynonymous (dN; amino acid altering)-to-synonymous (dS; silent changes) rate ratio (ω = dN/dS) under different models within a phylogenetic context. A signal of positive Darwinian selection can be detected when sites with sites with ω > 1 are present (Nielsen and Yang 1998; Swanson et al. 2000; Yang et al. 2000; Swanson et al. 2002).
The results presented here provide clear statistical evidence for adaptive molecular evolution of scorpion α- and β-toxin genes and thus support our hypothesis. Some functionally important sites identified previously in α-toxins are also predicted by our analysis.
Materials and Methods
Sequence Data and Tree Construction
Published cDNA sequences of α- and β-toxins were used in our study, which include 19 α-toxins from five species (Old World scorpions) and 15 β-toxins from two species (New World scorpions). Toxin sequences with some gaps were not included in this analysis. Insect sodium channel specific toxins (depressant and excitatory types) are excluded due to the lack of sufficient sequence data. Only codons encoding amino acids in mature toxins with the signal peptide and C-terminal extra residue removed were used. Considering that large gaps must be introduced when aligning α- and β-toxins together, which may significantly reduce maximum likelihood (ML) power, we separated them into two alignments and tested them respectively for positive selection. To construct trees that have been used for the estimation of dN/dS by the ML method, the amino acid sequences were aligned using CLUSTALW (http://www2.ebi.ac.uk/clustalw/ ) (Figs. 1 and 2). The nucleotide sequences were aligned accordingly. ML (Phylip 3.5) and NJ (neighbor-joining) methods (MGEA 2.1) were used to obtain tree topologies for further analysis. To get insight into the evolutionary relationship between α- and β-toxins, several trees were also made based on the combined nucleotide sequence alignment by using different methods implemented in MGEA 2.1 (http://www. megasoftware.net ).
To test for positive selection at single sites of aligned scorpion toxin sequences, we performed a maximum likelihood analysis to estimate the nonsynonymous-to-synonymous rate ratio (ω = dN/dS) using the CODEML program of the PAML software package (http://abacus.gene.ucl.ac.uk/software/paml.html). The presence of a positively selected rate class is detected by comparing the likelihood of a neutral model with that of a selection model (Yang and Bielawski 2000; Yang 2002). The neutral models constrain ω for each amino acid site between 0 and 1, where ω = 0 corresponds to purifying selection (selection acting against deleterious mutations) and ω = 1 to neutral evolution, whereas the selection models additionally allow positively selected (ω > 1) rate classes. We used six different models for ω ratio distribution among sites (Anisimova et al. 2001; Bielawski and Yang 2003). The M0 model (one-ratio) assumes that all sites have the same ω ratio. The M1 model (neutral) assumes two classes of sites in proteins: the conserved sites (ω = 0) and the neutral sites (ω = 1). The M2 model (selection) adds a third class of sites with ω as a free parameter, thus allowing for sites with ω > 1. Model M3 (discrete) uses a general discrete distribution with three site classes, with the proportions p0, p1, and p2 and the ω ratios ω0, ω1, and ω2. The M7 model (beta) allows sites to have 10 different ω ratios in the interval (0, 1), which are calculated from the beta distribution with parameters p and q. Model M8 (beta and ω) adds an extra class of sites to the beta (M7) model and allows the sites to have ω > 1. Regarding the parameter-rich models (M7 and M8), the calculations were run twice as recommended by the author (http://abacus.gene.ucl.ac.uk/software/paml.html ). once with initial ω > 1 and again with ω < 1. The results corresponding to the higher likelihood value are used.
From the above models, we constructed three likelihood ratio tests (LRTs) which compare a model that does not allow for positive selection with a model that allows for positive selection (M0/M3, M1/M2, and M7/M8). After ML estimates of parameters were obtained, the empirical Bayesian approach is used to calculate the probability that a specific site belongs to a given rate class depending on the data at that site (Nielsen and Yang 1998; Yang et al. 2000). Anisimova et al. have indicated that the power and accuracy of the Bayesian prediction are particularly influenced by the divergent degree and numbers of the sequences analyzed. To obtain a reliable result, the authors recommended using multiple models to identify sites under positive selection (Anisimova et al. 2002). Regarding our analysis, all the three models (M2, M3, and M8) pinpoint almost-identical sites in both α- and β-toxin genes, suggesting the feasibility of this prediction. Sites with a high probability (≥90%) of coming from the class with ω > 1 are likely to be under positive selection and were mapped onto the structures using the Chimera program (http://www.cgl.ucsf.edu/chimera/ ).
Evolution of Scorpion α- and β-toxin genes
Both scorpion α- and β-toxins are the members of duplicated gene families, and conserved gene structure and three-dimensional (3D) folds have suggested a common ancestor for them (Becerril et al. 1993; Froy et al. 1999). To get insight into their evolutionary relationship, we constructed several trees from their DNA sequences. Sequence alignment was carried out based on toxin structural identity, in which some gaps were removed. The aligned sequences were used to make trees using several different methods implemented in MGEA 2.1 (e.g., UPGMA, neighbor-joining [NJ], and minimum evolution [ME]). All these methods produced similar results, in which the α- and β-groups cluster into two separate classes (supported by high bootstrap values) (Fig. 3). Because of the recent work reported by Gordon et al., in which the authors identified the first typical β-toxin (Lqhβ1) from an Old World scorpion and demonstrated that one scorpion species can produce α- as well as β-toxins (Gordon et al. 2003), we can now hypothesize that an early gene duplication event led to the formation of ancestral α- and β-toxins which subsequently diverged into two distinct gene clusters by duplication followed by positive selection (Fig. 3). Thus, α- and β-toxins form a paralogous not an orthologous gene family. Despite the fact that most of the α-toxins were isolated from Old World scorpions, while β-toxins are mainly derived from New World scorpions (Possani et al. 1999), the finding of Lqhβ1 provides evidence supporting the early gene duplication event occurring before, not after, the separation of the continents (Gordon et al. 2003).
Test of Positive Selection and Identification of Corresponding Sites
To test positive selection of scorpion α- and β-toxins, we implemented six models (M0, M1, M2, M3, M7, and M8), which allowed us to construct three LRTs. For the α-toxin gene, parameter estimates from the models suggest that the one-ratio model (M0) fits the data worse than any other model (Table 1). Hence, it is rejected that all sites have the same ω ratio in the scorpion α-toxin gene. Three selection models (M2, M3, and M8) do detect the existence of a substantial proportion of positively selected sites. Especially, these models suggest a similar proportion (0.28–0.29) of sites under positive selection with a similar ω (2.76–3.04). LRTs indicate that these models significantly increase the likelihood scores compared with models with no positive selection. For example, the beta model (M7) does not allow for sites with ω > 1. The selection model (M8) adds an additional site class, with the ω ratio estimated to be 2.76. The likelihood value under this model is −1573. The LRT statistic (2Δl) is 24.76, which is much greater than the critical value from a χ2 distribution with df = 2 (M8 has two more parameters than M7). This suggests that M8 fits the data better than M7 and thus indicates the existence of positively selected sites with ω > 1. Two additional LRTs performed by comparing the 2Δl values between M0/M3 and M1/M2 led to consistent results (Table 1). Thus, our analysis provides unequivocal statistical evidence to support an adaptive molecular evolution occurring with scorpion α-toxins. Similarly, for the β-toxin gene, parameter estimates and LRTs also suggest the presence of sites under positive selection (Table 2).
Parameter estimates and likelihood ratio statistics (2Δl) for the scorpion α-toxin gene
Estimates of parameters
Positively selected sites
M0 (one ratio)
ω = 1.03
p0 = 0.26, ω0 = 0.03
10, 17, 18, 32, 37, 38, 39, 41, 54, 56, 60
p1= 0.45, ω1 = 0.95
p2= 0.28, ω2 = 2.78
p0= 0.24, ω0 = 0
p1= 0.76, ω1 = 1
p0= 0.22, ω0 = 0
10, 17, 18, 32, 37, 38, 39, 41, 54, 56, 60
p1= 0.49, ω1 = 1
p2= 0.29, ω2 = 3.04
p= 0.15, q = 0.08
M8 (β & ω)
p1= 0.29, ω = 2.76
10, 17, 18, 32, 37, 38, 39, 41, 54, 56, 60
p0 = 0.71
p= 0.09, q = 0.06
Parameter estimates and likelihood ratio statistics (2Δl) for the scorpion β-toxin gene
Estimates of parameters
Positively selected sites
M0 (one ratio)
ω = 0.61
p0 = 0.29, ω0 = 0.00
9, 10, 17, 18, 24, 27, 31, 56, 64
p1= 0.46, ω1 = 0.54
p2= 0.24, ω2 = 2.82
p0= 0.35, ω0 = 0
p1= 0.65, ω1 = 1
p0= 0.33, ω0 = 0
9, 10, 17, 18, 24, 31, 56, 64
p1= 0.49, ω1 = 1
p2= 0.18, ω2 = 4.26
p= 0.21, q = 0.21
M8 (β & ω)
p1= 0.20, ω = 3.24
9, 10, 17, 18, 24, 31, 56, 64
p0 = 0.80
p= 0.27, q = 0.37
To identify these sites, the Bayesian approach was used to calculate the posterior probabilities of ω classes for each site. Significantly, all selection models did identify seven identical positively selected sites with p > 0.95 (ω > 1) and four with p > 0.9 (ω > 1) in α-toxins (Table 1). For the β-toxins, M2 and M8 produced consistent predictions, which included six sites with p > 0.95 (ω > 1) and two with p > 0.9 (ω > 1). The discrete model (M3) identified similar positively selected sites with M2 and M8 with only one exception: site 27 (Table 2).
Location of Positively Selected Sites and Comparison with Functional Epitopes
Given that sites identified under positive selection have offered a selective advantage for gain of new functions along evolution (Swanson and Vacquier 2002), it is possible to hypothesize that these identified sites might be involved in toxin and sodium channel recognition and play an important role in functional diversification of toxins. The sites identified under positive selection are mapped onto BmK M1, being a representative for α-toxins, and CsEv2 for β-toxins based on two opposite molecular surfaces (faces A and B) (Figs. 4A and B) (He et al. 1996).
For the α-toxins, 9 sites of 11 are exposed on the molecular surface with >‰30% accessibility calculated by the Swiss-PDBViewer program (http://us.expasy.org/spdbv/ ). Most of the residues identified avoid secondary structures and are situated on the outer surface of the α/β scaffold. Only residues 37 and 38 are located in a β-sheet. Residues 10, 17, 18, 54, and 56 are visible on face A, where they form two clusters located on two sides of face A (Li et al. 1996; He et al. 1999). After a rotation of 180° along the Y-axis, residues 32, 37, 38, 39, 41, and 60 all emerge on the right side of face B with only one exception: residue 32 (Fig. 4A). For the β-toxins, four of eight residues have more than 30% accessibility and two (sites 24 and 31) are located in the α-helix. Remarkably, all eight positively selection sites are concentrated in face A.
The putative bioactive surface of LqhαIT, an α-toxin isolated from the scorpion Leiurus quinquestriatus hebraeus, has been elucidated by site-directed mutagenesis (Zilberberg et al. 1997; Gurevitz et al. 1998). Moreover, one loop of bukatoxin composed of residues PDKVP has also been suggested to possibly be involved in the interaction of scorpion α-toxins with sodium channels (Srinivasan et al. 2001). Therefore, these data allow us to compare the positively selected sites with the functionally important regions. Of 11 sites identified here, 6 have been found in the functional region, sites 10, 17, 18, 38, 54, and 56 (Gordon and Gurevitz 2003). In addition, a region of the B-loop (sites 37–43), which has been suggested to be important for the animal group preference (insects versus mammals), is also identified to contain sites under positive selection (sites 37–39 and 41). The functional sites for β-toxins have not been characterized so far, so similar comparisons cannot be performed. However, sequence alignment based on structural identity highlights four homologous positively selected sites between α- and β-toxins (Fig. 4C). All these sites have been characterized to be functionally crucial to α-toxins. In β-toxins, these sites combined with the other four sites found by our analysis form an integral surface located in face B (Fig. 4B) and are putatively involved in channel recognition. Comparison of positively selected sites of α- and β-toxins reveals an obvious difference at the B-loop. In α-toxins, the positively selected sites in this loop form one cluster on face B, but β-toxins lack this region (Fig. 4).
Gene duplication leading to new gene functions (functional divergence) is a common evolutionary event at the molecular level (Zhang et al. 1998; Long 2001; Yang et al. 2002). Two hypotheses have been proposed to account for the reasons responsible for the functional divergence. Neutral theory insists that random fixation of neutral mutations plays a major role in the evolution of new gene functions due to relaxation of selective constraints and environmental changes (Kimura 1983). In contrast, the selection theory emphasizes that positive Darwinian selection accelerates the fixation of advantageous mutations (largely nonsynonymous mutations), and as a consequence the functional divergence occurs (Yang and Bielawski 2000). When an elevated ω ratio does not exceed 1, the two theories are often difficult to distinguish. However, the evidence becomes convincing for positive selection if the ω ratio is more than 1 (Yang and Bielawski 2000). Based on this criterion, some duplication genes have been suggested to be subjected to positive selection (Yang and Bielawski 2000; Wolfe and Li 2003).
It is known that a large number of animal toxins are encoded by multigene families and each gene family exhibits a variety of diverse functions (Kini and Chan 1999; Kordis and Gubensek 2000). Hence, they provide excellent examples to address the debate between the neutral and the selection theories (Li 1997). By comparing the rates of nonsynonymous and synonymous substitutions, adaptive molecular evolution driven by positive selection for several animal toxin gene families has been suggested, of which only one invertebrate example is included (Kini and Chan 1999; Kordis and Gubensak 2000; Conticello et al. 2001).
By using a new and more robust statistical analysis (the maximum likelihood method) to test for positive selection, we provide unequivocal statistical evidence for the adaptive evolution of scorpion sodium channel toxins (α- and β-groups) (Nielsen and Yang 1998; Yang et al. 2000). This is shown to be a clear example of gene duplication followed by positive selection in invertebrate toxins. Gene duplication following the divergence of α- and β-toxins gave rise to α- and β-toxin families. In these evolutionary events, positive selection played a major role for their functional divergence.
What forces have driven the adaptive evolution of α- and β-toxins? One force may be derived from their targets, namely, VGSCs. Because sodium channels have many subtypes and are distributed in different tissues (cardiac and skeletal muscles, neurons), an accelerated evolution is required to expand their target range in order to be a more efficient “killer” and have a better defense against predators. Alternatively, due to environmental changes, prey might become extinct or less available in the region they live. To be able to adapt to this challenge, positive selection of toxins (for targeting different sites present in new prey) could be essential for the survival of certain scorpion species.
Some positively selected sites whose amino acid changes can provide fitness for toxin functions are found to be located in the functional surface of the studied toxins, indicating that they are important for toxin–sodium channel interaction. Previous studies have suggested that the functional diversification of scorpion sodium channel toxins might have been achieved along evolution via structural reconfiguration of the C-tail (Gurevitz et al. 2001). Here, we address additional possibilities in which the importance of the B-loop in functional diversification of α- and β-toxins is highlighted. The importance of this region for animal preference has been suggested but most of its residues are still not elucidated as active sites except site 38, which has been included in the functional region of LqhαIT (Gordon and Gurevitz 2003). As shown in Fig. 4, α- and β-toxins share similar evolutionary epitopes in face A but obvious differences can be found in face B, where four positively selected sites constitute the major evolutionary epitopes of α-toxins. Remarkably, this epitope is completely lacking in β-toxins. One corollary from this is that the B-loop extension and subsequent adaptive evolution is a major event by which α-toxins gained accessibility to target receptor 3.
The nucleotide sequence alignments of scorpion α- and β-toxins which were used in our calculations are available from the corresponding author.
We would like to thank Prof. Ziheng Yang for helping us with the maximum likelihood methods. This work was partly supported by a K.U.Leuven postdoctoral fellowship to S.Z.