Recruitment and diversification of an ecdysozoan family of neuropeptide hormones for black widow spider venom expression

Abstract

Venoms have attracted enormous attention because of their potent physiological effects and dynamic evolution, including the convergent recruitment of homologous genes for venom expression. Here we provide novel evidence for the recruitment of genes from the Crustacean Hyperglycemic Hormone (CHH) and arthropod Ion Transport Peptide (ITP) superfamily for venom expression in black widow spiders. We characterized latrodectin peptides from venom gland cDNAs from the Western black widow spider (Latrodectus hesperus), the brown widow (L. geometricus) and cupboard spider (Steatoda grossa). Phylogenetic analyses of these sequences with homologs from other spider, scorpion and wasp venom cDNAs, as well as CHH/ITP neuropeptides, show latrodectins as derived members of the CHH/ITP superfamily. These analyses suggest that CHH/ITP homologs are more widespread in spider venoms, and were recruited for venom expression in two additional arthropod lineages. We also found that the latrodectin 2 gene and nearly all CHH/ITP genes include a phase 2 intron in the same position, supporting latrodectin’s placement within the CHH/ITP superfamily. Evolutionary analyses of latrodectins suggest episodes of positive selection along some sequence lineages, and positive and purifying selection on specific codons, supporting its functional importance in widow venom. We consider how this improved understanding of latrodectin evolution informs functional hypotheses regarding its role in black widow venom as well as its potential convergent recruitment for venom expression across arthropods.

1. Introduction

Venoms are protein-rich secretions that have evolved in predatory animals for the purpose of prey immobilization and defense (Casewell et al., 2013; Fry et al., 2009). Venom production has independently arisen in several animal lineages including cnidarians, spiders, myriapods, scorpions, cone snails, cephalopods, snakes and mammals (Fry et al., 2009; Kordis and Gubensek, 2000; Wong and Belov, 2012). The venom produced by each species is often a complex cocktail of protein neurotoxins, hemotoxins and proteases (Escoubas et al., 2006; Fry et al., 2009; Sollod et al., 2005). Moreover, it appears that homologous genes were convergently recruited for venom expression in divergent taxa (Casewell et al., 2013; Fry et al., 2009). Many venom toxins originate from a gene duplicate encoding a structurally stable, cysteine-rich protein involved in a rapidly acting physiological process (Fry et al., 2009). If such gene duplicates experience relaxed selection, mutations may accumulate in their coding and regulatory regions, causing them to be expressed in venom tissue. Further duplication of venom-expressed genes, coupled with high mutation rates, generates a multigene venom toxin family targeting a variety of receptors in diverse prey species (Duda et al., 2009; Fry et al., 2009; Sollod et al., 2005).

Numerous studies have focused on the activity and composition of black widow spider venom (genus Latrodectus), but few have considered its evolution (Garb and Hayashi, 2013; Holz and Habener, 1998; Ushkaryov et al., 2004). Black widow spider venom has a potent neurotoxic effect on mammals, and bite symptoms may include nausea, vomiting and intense pain (Grishin, 1998; Nicholson and Graudins, 2002; Ushkaryov et al., 2004; Vetter and Isbister, 2008). Latrodectus venom has largely been characterized from the Eurasian species L. tredecimguttatus, though the genus contains 31 species (Platnick, 2013). L. tredecimguttatus’ venom is dominated by latrotoxins, which are large polypeptides ~1200 amino acids long (Ushkaryov et al., 2004). Of the four latrotoxins, α-latrotoxin (α-LTX) is the only vertebrate neurotoxin and is responsible for the effects associated with widow bites (Ushkaryov et al., 2004). α-LTX acts as a calcium ion channel in the presynaptic nerve terminal membrane and causes massive neurotransmitter release (Orlova et al., 2000; Ushkaryov et al., 2004).

Latrodectins or α-latrotoxin associated Low Molecular Weight Proteins (α-latrotoxin LMWPs), are a second family of venom peptides from L. tredecimguttatus venom, only known from two cDNA sequences (Kiyatkin et al., 1992; Pescatori et al., 1995). Latrodectins are peptides of ~70 amino acids that cannot be separated from latrotoxins using standard protein purification (Kiyatkin et al., 1992, 1990; Pescatori et al., 1995; Volkova et al., 1995). Multiple studies have demonstrated that purified latrodectin is not toxic in insects and mammals (Gasparini et al., 1994; Grishin et al., 1993; Kiyatkin et al., 1995; Volkova et al., 1995). However, latrodectins may function as subunits of a latrotoxin complex (Kiyatkin et al., 1992), even though latrotoxins do not require latrodectins for neurotransmitter release (Dulubova et al., 1996; Grishin et al., 1993; Kiyatkin et al., 1995; Volynski et al., 1999).

Gasparini et al. (1994) noted that latrodectins have sequence similarities to the Crustacean Hyperglycemic Hormone (CHH) family, which contains neuropeptides from crustaceans that includes Type I peptides involved in ionic metabolism and osmoregulation and Type II peptides, comprising more specialized developmental hormones (Montagne et al., 2010). The CHH family exists in insects as the Ion Transport Peptides (ITPs), and CHH/ITP homologs have also been identified in ticks and nematodes (Montagne et al., 2010). The latrodectins, CHHs, and ITPs are similar in length, share six conserved cysteines in the mature peptide that adopt the same disulfide bond pairing, and have a similar alpha-helical structure (Gasparini et al., 1994). It is likely that latrodectins were recruited for venom gland expression from a broadly expressed spider CHH/ITP homolog. However, the diversity of latrodectins or their relationships to the CHH/ITP neuropeptide superfamily has not been explored in a phylogenetic framework.

We investigated the expression and evolution of latrodectin sequences across widow spiders using venom gland cDNA libraries from the Western black widow spider (L. hesperus), the brown widow spider (L. geometricus), and the cupboard spider (Steatoda grossa), in the putative sister genus to Latrodectus (Agnarsson, 2004; Arnedo et al., 2004). We examined these sequences with putative homologs identified from database searches using phylogenetic and molecular evolutionary analyses to determine patterns of selection on and diversification among latrodectins. We also characterized the partial structure of latrodectin genes, which provides novel support for their derivation from CHH/ITP neuropeptides. Our results advance understanding of the evolutionary origins and diversity of venom proteins, as well as the function of latrodectins in black widow venom.

2. Materials and Methods

2.1 cDNA library construction and screening

L. hesperus and L. geometricus were collected in California (Riverside and San Diego, respectively). S. grossa were purchased from SpiderPharm (Yarnell, Arizona). 42 L. hesperus, 27 L. geometricus, and 25 S. grossa adult females were used to make separate venom gland cDNA libraries from each species. Total RNA was extracted from homogenized venom glands using Trizol™ and purified using the RNeasy Kit (Qiagen Inc., Valencia, CA). mRNA was isolated from total RNA using the Dynabeads™ mRNA purification kit (Invitrogen Corp., Carlsbad, CA). cDNA was synthesized using the protocol in Garb and Hayashi (2005). cDNAs were size-selected for transcripts ≥1000 bp in length with a Chromaspin 1000 column. This retains many fragments <1000 bp, but reduces their overrepresentation. cDNAs were ligated in the pZErO™-2 plasmid (Invitrogen Corp). Top10 E. coli was transformed with recombinant plasmids via electroporation and grown on agar plates. Clones were arrayed on 12-18 96-well microplates, which were screened for cDNA size using electrophoresis (Garb and Hayashi, 2005).

2.2 cDNA sequencing and bioinformatic analyses

cDNA inserts 200 bp were sequenced using the universal SP6 primer. Latrodectin cDNAs were also sequenced in the reverse direction using the T7 primer. Sequences were edited using Sequencher 4.8 (Gene Codes, Ann Arbor, MI), and assembled into contiguous sequences (contigs). Highly similar sequences were identified using the BLASTclust program (http://toolkit.tuebingen.mpg.de/blastclust/). All cDNAs were used as BLASTx queries against the NCBI nr protein database (Altschul et al., 1990). Latrodectin cDNAs were translated based on the predicted frame of the top BLASTx hit. A sampling of CHH/ITP homologs from arthropod and nematode taxa were included in subsequent analyses (Table 1) from phylogenetically distinct lineages of the CHH/ITP superfamily as determined by Montagne et al. (2010). Following Montagne et al. (2010), phylogenetic analyses across the CHH/ITP family were restricted to sequences representing the short, amidated forms of these peptides. Additional long forms of CHH/ITP peptides exist in crustaceans and insects, which are the result of 1-2 additional exons that are alternatively spliced. These additional exons appear to be unique to Pancrustacea and present problems for phylogenetic analyses with sequences from other taxa, as these additional exons do not exist in arachnid and nematode ITP homologs (Montagne et al. 2010). Additional arachnid latrodectin homologs were identified with tBLASTx searches of NCBI’s EST database (dbEST) using L. tredecimguattus latrodectins 1 and 2 as queries . Putative signal peptides in sequences were predicted using SignalP 4.0 (http://www.cbs.dtu.dk/services/SignalP/). Potential cysteine inhibitor knot motif structure was predicted with the Knoter1d program (http://knottin.cbs.cnrs.fr/Tools_1D.php). Sequences were submitted to the ClanTox (http://www.clantox.cs.huji.ac.il/index.php) (Naamati et al., 2009) classifier of animal toxins. Toxin prediction by ClanTox for the query sequence is highly dependent on the presence and distribution of cysteines, based on a training set of true and false ion channel toxin inhibitors (Naamati et al., 2009).

Table 1.

Sequences used in analyses with protein names and GenBank accession numbers.

Species	Protein Descriptor a.	Protein Accession	Nucleotide Accession
Sequences obtained from GenBank
Latrodectus tredecimguttatus	Latrodectin 1; α -latrotoxin-associated low molecular weight protein (LMWP)	CAA44830	X63116
Latrodectus tredecimguttatus	Latrodectin 2; α-latrotoxin-associated low molecular weight protein 2 (LMWP2)	AAY33774	DQ011856
Tegenaria agrestis	TaITX-1; U1-agatoxin-Ta1a; Paralytic insecticidal toxin 1	CAA11839	AJ224127
Tegenaria agrestis	TaITX-2; U1-agatoxin-Ta1b; Paralytic insecticidal toxin 2	CAA11840	AJ224128
Tegenaria agrestis	TaITX-3U1-agatoxin-Ta1c; Paralytic insecticidal toxin 3	CAA11841	AJ224129
Mesobuthus gibbosus	Venom cDNA clone Mg_AFT_30E12	Inferred from EST	CB334130.1
Loxosceles laeta	EY188575.1; LLAE0203C L. laeta venom cDNA	Inferred from EST	EY188575
Loxosceles intermedia	HO003697; EST1080 L. intermedia venom cDNA	Inferred from EST	HO003697
Dermacentor variabilis	ITP; Prepro ion transport-like peptide	ACC99599	EU620224
Ixodes scapularis	ITP; CHH (PO-type) variant 1 precursor	XP_002400720	XM_002400676
Caenorhabditis elegans	ITP1; Uncharacterized protein ZC168.2	ZC168.2	Z70312.1 positions 9773- 10008
Caenorhabditis elegans	ITP2; Protein C05E11.6	CCD63262	NM_076381
Daphnia magna	ITP; Putative ion transport peptide-like	ABO43963	EF178503
Aedes aegypti	ITP; Ion-transport peptide	AAY29661	AY950504
Drosophila melanogaster	ITP; Ion-transport peptide	ABZ88142	NM_001169822 positions 1080-1406
Schistocerca gregaria	ITP; Ion-transport peptide	AAB16822	U36919
Homarus gammarus	CHH; Hyperglycemic hormone B	ABA42180	DQ181792
Homarus gammarus	VIH; Vitellogenesis inhibiting hormone	ABA42181	DQ181793
Microctonus hyperodae	Venom protein 10	ABY19395	EU249359
Sequences characterized in this study
Latrodectus hesperus	LHV117		KF751506
Latrodectus hesperus	LHV218		KF751507
Latrodectus hesperus	LHV45		KF751508
Latrodectus hesperus	LHV319		KF751509
Latrodectus hesperus	LHV238		KF751510
Latrodectus hesperus	LHV229		KF751511
Latrodectus geometricus	LGV89		KF751512
L.geometricus	LGV361		KF751513
L.geometricus	LGV332		KF751514
L.geometricus	LGV382		KF751515
Steatoda grossa	SGV242		KF751516
Steatoda grossa	SGV282		KF751517
Steatoda grossa	SGV150		KF751518
Steatoda grossa	SGV152		KF751519
Steatoda grossa	SGV23		KF751520
Steatoda grossa	SGV335		KF751521
Steatoda grossa	SGV311		KF751522
Steatoda grossa	SGV81		KF751523
Steatoda grossa	SGV41		KF751524

^aAlternative protein names provided where available and separated by semicolons; protein descriptors derived from UniProtKB database or NCBI.

2.3 Latrodectin-2 gene characterization

An alignment of latrodectin cDNAs was used to develop primers to PCR-amplify latrodectin 1 and 2 from genomic DNA. Only primers for latrodectin 2 (LD2-F 5′-GATGCTTAAGCTTATCTGCATTG-3′ and LD2-R 5′-GGATATTGTGTAGTAAAGCAATTC-3′) produced amplifications from Latrodectus species. PCR fragments were ligated into the pCR 2.1 TOPO vector (Invitrogen Corp.) and electroporated into Top10 E. coli. Clones were selected for plasmid purification and sequenced with M13F and M13R primers. Intron position and sequence was inferred from interrupted coding sequence determined from cDNAs, and identifying putative 5′GU and AG 3′ splice sites.

2.4 Phylogenetic and molecular evolutionary analysis

Phylogenetic trees were constructed from the sequences listed in Table 1. Amino acid sequences were aligned using the MUSCLE program (Edgar, 2004), followed by manual adjustment. Amino acid alignments were used to guide an alignment of corresponding nucleotides using tranalign (http://emboss.bioinformatics.nl/cgi-bin/emboss/tranalign); only unique sequences were included in analyses. Phylogenetic trees were computed from the nucleotide alignment with maximum likelihood (ML) and Bayesian methods, using the substitution model determined with the Akaike Information Criterion (AIC) in jModelTest (Posada, 2008). ML trees were constructed using PAUP 4b10 (Swofford, 2003). A heuristic search was performed using 100 random taxon addition (RTA) replicates with gaps as missing data. 100 bootstrap replicates were performed using three RTAs per replicate. Bayesian trees were estimated with Mr.Bayes 3.2.1 (Ronquist et al., 2012), using the model selected by the AIC by MrModeltest 2.3 (Nylander, 2004). The program was run for 5×10⁶ generations; trees were sampled every 1,000 generations. The first 25% of trees were removed as “burn-in” after standard deviation of split frequencies fell below 0.01. The post burn-in trees were used to construct a 50% majority-rule consensus tree, from which all Clade Posterior Probability (CPP) values were computed. The two nematode (C. elegans) sequences were used to root the tree, which represent basal sequences in the CHH/ITP superfamily (Montagne et al., 2010).

Tests of selection on codons and along branches of the phylogeny were performed using the codeml package of PAML 4.3 (Yang, 2007) and the HyPhy package via the www.datamonkey.org web server (Delport et al., 2010). Estimates of nonsynonymous substitutions per nonsynonymous site (d_N) over synonymous substitutions per synonymous site (d_S), or ω, for branches and sites in the latrodectin phylogeny, as well as for pairwise sequence comparisons were determined using codeml. These analyses were based on an ML phylogeny restricted to a clade of spider latrodectin sequences, and limited to single representatives from clades of nearly identical sequences, and all analyses used the cleandata=1 option (Yang, 2007). In codeml the MO, M1a, M2a, M3, M7 and M8 sites models were used to estimate ω under the assumption of no site variation (M0) in comparison to variable ω among sites (M3); and to compare models to test for positive selection (M1a vs. M2a and M7 vs. M8; Yang et al., 2000). We also tested for variable ω across branches, comparing the free-ratio to the one-ratio model, as well as the free-ratio model in comparison to when ω was fixed to 1. Statistical differences among models were estimated using a likelihood ratio test (LRT). We also used the HyPhy package in the www.datamonkey.org server (Delport et al., 2010) to perform the single-likelihood ancestor counting (SLAC) method, the fixed effects likelihood (FEL) method, and the Fast Unbiased Bayesian AppRoximation (FUBAR) method to detect positive and negative selection on codons, as well as the Mixed Effects Model of Episodic Diversifying Selection (MEME) method to detect diversifying episodic selection on codons. The GA-Branch module in datamonkey was used to fit d_N/d_S rate classes to branches of the tree. Analyses used the nucleotide evolution model selected by the AIC in the datamonkey model selection tool, the ML tree from the corresponding alignment, and were conducted using default significance levels.

3. Results

3.1 Sequence features and variability of latrodectins and their homologs

We sequenced 1,002 venom gland cDNAs from L. hesperus (355 ESTs), L. geometricus (281), and S. grossa (366). BLASTx results revealed 30 cDNAs were most similar to latrodectins from L. tredecimguttatus (7 from L. hesperus, 11 from L. geometricus and 12 from S. grossa). Nineteen of the 30 sequences were unique at the nucleotide level (6 from L. hesperus, 4 from L. geometricus and 9 from S. grossa; Table 1). Both L. hesperus and L. geometricus had latrodectins that clustered into two groups based on BLASTclust results, with the two groups being most similar to latrodectin 1 or 2 from L. tredecimguttatus. S. grossa sequences contained three distinct latrodectin clusters, suggesting the presence of three paralogs. Across these species, translated latrodectins ranged in length from 88 to 97 amino acids, and shared a minimum of 22.9% pair-wise identity, with an average of 41.4% pair-wise identity.

After the top BLASTx hits of latrodectin 1 or 2 from L. tredecimguttatus, BLASTx searches with latrodectin cDNAs included many hits to Crustacean Hyperglycemic Hormone (CHH) peptides and Ion Transport Peptides (ITP). Translated BLAST searches of NCBI’s dbEST database also identified hits to L. tredecimguattus latrodectins in four other arachnid species. These included three insecticidal venom neurotoxins TaITX 1-3 from the hobo spider Tegenaria agrestis (Family Agelenidae), venom cDNAs from recluse spiders Loxosceles laeta and Loxosceles intermedia (Family Sicariidae), and a venom gland EST from the scorpion Mesobuthus gibbosus. In addition, venom protein 10 from the wasp Microctonus hyperodae was also identified with PSIBLAST searches of NCBI’s nr database as having similarity to latrodectins (Table 1). All sequences used in phylogenetic analyses have six cysteines in their predicted mature peptides that are 100% conserved (Fig. 1a). All examined sequences, excepting the Daphnia magna putative ITP and the Schistocerca gregaria ITP, were predicted to have a signal peptide by SignalP (Fig. 1a). The Knoter1d program predicted that none of the examined sequences were inhibitor cysteine knot (ICK) toxins. The ClanTox server classified all Latrodectus and Steatoda mature latrodectins, as well as the Loxoceles, Tegenaria, and Mesobuthus homologs as “possibly toxin-like”, rather than “probably toxin-like” or “toxin-like”. All other CHH/ITP homologs were classified as “probably not toxin-like”.

Fig. 1. Alignment of latrodectin venom peptides with ecdysozoan CHH/ITP neuropeptide homologs and conserved intron position.

(a) Alignment of latrodectin venom peptides from Latrodectus and Steatoda with putative homologs from arthropod venom cDNAs and from the ecdysozoan CHH/ITP neuropeptide superfamily. Numbers 1-6 on top indicate six cysteine residues conserved across all sequences. Predicted signal peptides are underlined and dashes represent hypothesized insertions/deletions. Arrow indicates position of conserved phase 2 intron shared among sequences; boxed region detailed in part b. Additional signal sequence MHHQKQQQQQKQQGEAP found in the Schistocerca ITP peptide (but absent in all other sequences), is upstream of alignment position 1 and not shown. (b) (Top) Schematic of primary structure for CHH/ITP/latrodectin peptides marking approximate position of six conserved cysteine residues, the fourth cysteine and the following two residues, the boxed region corresponds to the boxed region in part a; (Middle) gene structure illustrating the conserved phase 2 intron for the L. hesperus latrodectin 2 peptide; (Bottom) gene structure of Ixodes ion transport peptide showing conserved phase 2 intron. Nucleotides in square brackets represent intronic sequence, only showing canonical splice sites.

3.2 Latrodectin gene structure

Genomic PCR products of latrodectin 2 from L. hesperus and L. geometricus (GenBank Accessions: KF751525-KF751526) contained two exons, the sequences of which agree with their corresponding cDNA sequences, excepting 1-2 nucleotide differences that likely reflect allelic differences, as cDNAs and genomic DNA were derived from different individuals. The two exons were interrupted by a phase 2 intron in the codon of the residue following the fourth cysteine in each mature peptide (Fig. 1b). A phase 2 intron interrupting the codon of the residue following the fourth cysteine is also found in all CHH/ITP family members, except ITP2 genes from Caenorhabditis and the ITP gene of Trichinella spiralis (Montagne et al., 2010), and some CHH/ITP members have more introns in addition to the one in this position (Montagne et al., 2010). It is possible there are additional introns in genomic latrodectin 2, as the genomic sequence we obtained was restricted to the first 71 out of 88 total residues. However, our cDNA sequences do not suggest the potential for alternatively spliced forms of latrodectin genes. The L. geometricus latrodectin 2 intron is 1777 bp in length, 241 bp longer than the intron in L. hesperus. Moreover, the L. geometricus intron sequence is extremely divergent from the intron in L. hesperus (50.4% nucleotide identity), whereas the exons share 82% identity.

3.3 Phylogenetic analyses of latrodectins and CHH/ITP homologs

Maximum likelihood (ML) and Bayesian tree searches were performed on a nucleotide alignment of latrodectins and CHH/ITP homologs in Table 1. Model selection for this alignment using the AIC in jModelTest selected the TPM3uf +I+G model of nucleotide substitution with model parameters as follows: [A-C] = 1.5771; [A-G] = 2.2915; [A-T] = 1.0000; [C-G] = 1.5771; [C-T] = 2.2915; freqA = 0.2861; freqC = 0.2256; freqG = 0.2419; freqT = 0.2464; I = 0.039; G = 3.6720. The AIC in MrModeltest 2.3 selected the GTR+I+G substitution model. The Bayesian consensus tree from this model contained a clade including all spider venom cDNAs ((CPP)=1.00, Bootstrap Support (BS)=67%)), which is sister to the scorpion venom cDNA (CPP=0.92; Fig. 2). The ML searches recovered a single tree of −ln L=8174.43986, which was identical in topology to the Bayesian tree in Fig. 2, but resolved the two unresolved nodes. Tick (Ixodes + Dermacentor) ITPs appear more closely related to crustacean and insect CHH and ITPs than to the arachnid venom cDNAs (CPP=0.56); however, this relationship did not have bootstrap support in the ML analysis. Within the spider clade, the latrodectin cDNAs from Latrodectus and Steatoda are united with limited support (CPP=0.63) and this clade is sister to a grouping of Loxoceles and Tegenaria venom cDNAs (CPP=0.86). These relationships, though identical, are not well supported in the ML tree. Two clades containing either latrodectin 1 and 2 were recovered, each with strong support (CPP=1.00 for both; latrodectins 1 BS=62%, latrodectins 2 BS=84%). Within the latrodectin 1 clade, the sequences are largely arranged based on expectations from species relationships (Fig. 3), with L. hesperus being more closely related to L. tredecimguttatus than it is to L. geometricus (Garb and Hayashi, 2013; Garb et al., 2004). However, the latrodectin 2 clade united L. geometricus and S. grossa sequences. An ML tree of the spider sequences alone, sampling one sequence per paralog per species, showed a topology identical to that seen for the Steatoda/Latrodectus sub-clades within the larger tree (−ln L=2870.97520; Fig. 4). The reduced tree showed a strongly supported clade of Steatoda and Latrodectus latrodectin paralogs including latrodectin 1 (BS=98%), and moderate support for a clade including latrodectin 2 orthologs and the third S. grossa paralog (BS=57%).

Fig. 2. Phylogenetic relationships of latrodectins and ecdysozoan CHH/ITP neuropeptide homologs.

Phylogram of Bayesian consensus tree from analysis of nucleotides encoding latrodectins and putative CHH/ITP superfamily peptides. Values above nodes indicate Clade Posterior Probability values (CPP). Numbers below nodes indicate maximum likelihood bootstrap values from 100 replicates. The tree is rooted with C. elegans homologs. Hatched lines indicate shortened branch for figure quality.

Fig. 3. Species relationships and divergence dates for spiders expressing latrodectins and homologous venom peptides.

Phylogenetic relationships for spider species from which latrodectin peptides or their homologs were obtained and analyzed in this study. Divergence dates of major lineages at nodes as estimated by Ayoub et al. (2007) in millions of years ago (MYA). Species relationships are summarized from species phylogenies in Coddington et al. (2004) and Garb and Hayashi (2013).

Fig. 4. Branch-specific patterns of selection on the spider phylogeny of latrodectins and homologous spider venom peptides.

Maximum likelihood phylogenetic tree of latrodectin sequences and homologous spider venom peptides. Numbers above branches show ML estimates of ω (nonsynonymous/synonymous substitution rates) determined from the branch free-ratio model of codeml in PAML. Values of infinity are appended with estimates of nonsynonymous differences over synonymous differences for that branch. Numbers under the branches indicate ML bootstrap values for that node from 100 replicates.

3.4 Analyses of selection on latrodectins

Codeml estimates of ω across branches in the reduced latrodectin tree using the free ratio model ranged from 0.0034 to infinity (∞), the latter value being for branches where 0 synonymous substitutions were estimated (Fig. 4). Several of the branches where ω=∞ had estimates of relatively high numbers of nonsynonymous substitutions (e.g., 14.6, 32.3, 19.4, 5.1; Fig. 4), and some branches had ω values exceeding 1 (e.g., 13.6, 4.1; Fig. 4). The one ratio model estimated an overall ω value for all branches as 0.3597, and this model was a significantly worse fit than the free ratio model (Table 2). However, the one ratio model was a significantly better fit to the data than fixing ω to 1. This suggests an overall pattern of purifying selection on latrodectin lineages over time, with potential evidence for positive selection on several branches where ω exceeds 1. LRT comparisons rejected the M0 over the M3 model, suggesting variable ω across sites. However, the M2a model was not significantly different from M1a, nor was the M8 model significantly different from M7, and the Bayes Empirical Bayes (BEB) procedure found no evidence of positive selection on specific sites within latrodectin (i.e., Pr (ω >1)=0.90 or higher; Table 2). HyPhy analyses of positive selection on latrodectin codons identified seven positively selected sites, all of which were detected with the MEME method, which identifies episodic positive selection. Two of these sites were also significant for the FEL method analysis (Table 3). Two of the seven positively selected sites are located in the predicted signal peptide region, while the rest are equally distributed along the mature peptide. Fourteen codons were identified as negatively selected with significance in at least one HyPhy analysis, including five of the six conserved cysteine residues (Table 3). The average d_N/d_S value over the entire alignment estimated by the SLAC method was 0.475. The GA-Branch analysis fit all branches in the tree to two d_N/d_S rate classes: d_N/d_S of 0.257 to 66% of the branches and d_N/d_S of 0.867 to 34% of the branches. While the branch rate classes do not exceed 1, this result is consistent with variable rates of selection across the tree.

Table 2.

Codon substitution models for spider latrodectin ML tree (Fig. 4) with ω estimates and probabilities.

	Model	−ln L	Parameter estimates	χ 2	df	P
Branches	Free ratio	−1893.66
	One ratio	−1913.90	ω = 0.3597	40.49	21	0.0064
	Fixed ratio	−1937.048	ω =1	46.29	1	<0.0001
Sites	M0	−1913.90	ω =0.3597
	M3	−1877.97	p(0,1,2): 0.1061 0.5314 0.3625 ω(0,1,2): 0.0060 0.2675 0.8830	71.86	4	<0.0001
	M1a	− 1887.75	p(0,1): 0.5740 0.4261 ω(0,1): w: 0.2122 1.0000
	M2a	− 1887.75	p(0,1,2): 0.5740 0.3680 0.0580 ω(0,1,2): 0.2122 1.0000 1.0000	0	0	1
	M7	− 1882.28	p=0.8329 q= 1.0868
	M8	− 1882.03 p0= 0.9085	p= 0.9452 q= 1.5070 (p1= 0.0916) ω = 1.2587	0.51	2	0.7765

Table 3.

Summary of results from HyPhy analyses of codon-specific selection in the latrodectin alignment.

Codon a	SLAC dN-dS	SLAC p-valueb	FEL dN-dS	FEL p-valuec	FUBAR dN-dS	FUBAR Post. Pr. d	MEME ω+ f	MEME p-valuee
Positively selected codons
2	1.155	0.157	0.269	0.030	0.97	0.884	>100	0.041
7	0.040	0.655	0.164	0.089	−0.016	0.634	>100	0.086
12	−0.002	0.650	0.301	0.353	0.118	0.755	31.863	0.033
24	−0.035	0.664	−0.209	0.453	−1.352	0.275	>100	0.048
31	−0.376	0.794	−0.233	0.523	−1.243	0.265	>100	0.022
51	0.448	0.505	−0.017	0.939	−0.070	0.477	94.614	0.053
54	0.076	0.622	−0.150	0.684	−0.201	0.456	>100	0.062
Negatively selected codons
4	−0.088	0.583	−0.594	0.068	−0.225	0.761	-	-
15	−0.576	0.331	−0.286	0.110	−0.427	0.937	-	-
17	−0.592	0.347	−3.026	0.004	−3.339	0.980	-	-
19 (C-1)	−0.737	0.183	−0.074	0.084	−0.121	0.959	-	-
22	−1.239	0.125	−0.313	0.075	−0.379	0.894	-	-
23	−0.800	0.318	−1.178	0.057	−1.649	0.865	-	-
25	−1.680	0.073	−0.383	0.044	−0.438	0.915	-	-
26	−1.645	0.080	−0.965	0.026	−0.657	0.898	-	-
33 (C-2)	−2.272	0.006	−1.514	0.001	−1.934	0.998	-	-
36 (C-3)	−2.515	0.002	−0.597	0.000	−1.414	0.999	-	-
47 (C-5)	−1.298	0.041	−0.289	0.006	−0.487	0.993	-	-
48	−1.223	0.079	−0.197	0.073	−0.275	0.951	-	-
56 (C-6)	−1.163	0.048	−345.342	0.003	−2.134	0.995	-	-
60	−1.227	0.056	−0.223	0.027	−0.134	0.948	-	-

^a. Letter C in parentheses, followed by number indicates conserved cysteine according to positions defined in Fig. 1; note that C-4 has the same codon for all sequences in this alignment, as well as in larger alignment excepting the Microctonus and one C. elegans ITP sequence.

^{b-d, e}. Statistical support for positive or negative selection by each method is indicated by bold p-values or bold posterior probability values (Post. Pr.);

^f. values indicate inferred ω (β⁺/α), values where they exceed 100 include sites where α is 0.

4. Discussion

4.1 Multiple recruitments of CHH/ITP neuropeptides for venom gland expression

A particularly fascinating aspect of venom evolution is the observation that many similar proteins and peptides are expressed in the venom of animal lineages that have independently evolved venom production (Casewell et al., 2013; Fry et al., 2009; Wong and Belov, 2012). These multiple recruitments suggest that certain classes of proteins that are largely expressed in another body tissue are more likely to be recruited and retained for venom expression, and that while the recruitment event is convergent, the proteins involved may be homologous at some level, though they likely involve paralogs. In this study we provide an example of a widespread family of arthropod peptides, the CHH/ITP neuropeptides, which may represent another example of convergent recruitment for venom gland expression. Our finding of a phase 2 intron in the latrodectin codon following the fourth cysteine codon in its mature peptide, which is found in nearly all other CHH/ITP genes, reinforces the claim that latrodectins are derived from this ecdysozoan family of neuropeptide hormones (Gasparini et al., 1994). CHH/ITP neuropeptides share some of the characteristics of other protein families that have been independently recruited for venom expression, such as a signal peptide and multiple cysteine residues that participate in disulfide bonds (Gasparini et al., 1994).

The functions of CHH/ITP neuropeptides are varied, including roles in ionic metabolism, regulation of molting and reproduction, and osmoregulation (Montagne et al., 2010), and their widespread expression in various body tissues may further predispose them for venom recruitment. Recruitment of CHH/ITP neuropeptides for venom expression may have happened at least three times: once in Hymenoptera, once in scorpions, and at least once in spiders. Although the hymenopteran, scorpion and spider CHH/ITP/latrodectin peptides are grouped as a clade in our phylogenetic trees, we suggest as many as three venom recruitments, because each lineage independently evolved venom production. This is a parsimony-based argument, as each lineage is more closely related to non-venomous species, and because their venoms are produced in non-homologous glands (cheliceral glands in spiders vs. abdominal venom glands in scorpions and Hymenoptera). This scenario of three recruitment events can be tested in future studies by performing whole-genome sequencing and expression analyses of each group, to confirm whether genes encoding venom-specific peptides are derived from duplicates of CHH/ITP genes that are maintained for expression in other body tissues.

4.2 Widespread distribution of CHH/ITP/latrodectins in spider venoms

Previous sequences of the latrodectin/α-latrotoxin LMWP venom peptides were known from L. tredecimguttatus (Kiyatkin et al., 1992; Pescatori et al., 1995), and our results confirm these peptides are expressed in the venoms of other Latrodectus and Steatoda species. Our bioinformatic searches further suggest that latrodectin/α-latrotoxin LMWP homologs are also present in the venoms of two additional spider genera in very distantly related families: Tegenaria (Agelenidae) and Loxosceles (Sicariidae). Agelenidae, Sicariidae and Theridiidae (the family of Latrodectus and Steatoda), are representatives of three deep lineages (RTA clade, Haplogynae and Orbiculariae, respectively; Fig. 3) within the suborder Araneomorphae (true spiders) (Coddington et al., 2004). Assuming a single recruitment of CHH/ITP/latrodectins for venom expression in spiders, we estimate that this event took place at least 375 million years ago, the age of the common ancestor of these families (Ayoub et al., 2007). Further characterization of spider venoms is needed to determine the greater distribution of latrodectins, and to test whether a single ancient recruitment for venom expression was followed by numerous losses, or if multiple recruitments took place in spiders. For species having only venom cDNAs (but not cDNAs in the same protein family from other tissues) available for phylogenetic analyses, venom cDNAs are likely to form a monophyletic clade, potentially erroneously implying a single recruitment for venom expression (Casewell et al., 2012).

The denser sampling of latrodectin sequences we provide allows for more detailed evolutionary analyses. The phylogenetic arrangement of the Latrodectus and Steatoda sequences strongly supports the expression of two-three latrodectin paralogs in the venom of each species. Relationships within the latrodectin 1 clade are largely consistent with expected relationships for Latrodectus species, e.g., (((L. hesperus + L. tredecimguttatus) L. geometricus)) S.grossa))); (Garb and Hayashi, 2013; Garb et al., 2004). However, there is evidence for additional lineage-specific gene duplication within S. grossa. Given the available data, a lineage-specific duplication in Steatoda is more plausible than this third paralog arising in the common ancestor of all four species, but only being expressed in S. grossa. In contrast to the latrodectin 1 clade, the latrodectin 2 clade does not agree with expected phylogenetic relationships, as the S. grossa latrodectin 2 sequence is joined to the L. geometricus sequence, although by a much longer branch. While this relationship may be an artifact of long-branch attraction, there is also the possibility of additional paralogs in this clade that were not sampled from the other species.

Additional evolutionary analyses suggest evidence for positive and negative selection acting on particular latrodectin sequences and codons. Specifically, codeml analyses identified four branches in the latrodectin phylogeny where branch-specific estimates of ω exceeded 1, as well as four additional branches where only nonsynonymous substitutions (ranging from 5.1-32.3 substitutions) were estimated (Fig. 4). Branches where ω exceed 1 sometimes precede gene duplication events, but are also associated with nodes connecting orthologs. Both latrodectin 1 and 2 appear to have experienced positive selection in the most recent common ancestor of the Latrodectus mactans clade (L. geometricus and L. rhodesiensis form the geometricus clade, all other species form the mactans clade within Latrodectus) (Garb and Hayashi, 2013; Garb et al., 2004), which suggests a shift in its functional activity coincident with the origin of black widow species. Similar analyses indicated accelerated substitution rates in the vertebrate neurotoxin α-latrotoxin preceding the origin of the Latrodectus genus, as opposed to the mactans clade (Garb and Hayashi, 2013). The venoms of species from the mactans clade (which contains both the black widow and red-back spiders) exhibit greater toxicity relative to members of the geometricus clade as well as to species outside of Latrodectus (Graudins et al., 2012, 2002; Müller et al., 1992). This increased toxicity of black widow venom has largely been attributed to α-latrotoxin sequence identity and expression levels (Garb and Hayashi, 2013; Graudins et al., 2012), but venom proteins such as latrodectins that may work synergistically with α-latrotoxin could also play an important role in explaining venom functional variation.

While codeml analyses did not find statistical support for positive selection at specific codons, HyPhy analyses found some statistical support of positive selection on 7 sites (largely due to episodic diversifying selection), as well as negative selection on 14 sites (over a total of 62 codons considered when excluding gapped regions). This included support for negative selection at 5 of the 6 conserved cysteine residues, all of which are expected to be critical for maintaining the disulfide bonds that shape the overall fold of this peptide (Gasparini et al., 1994). Five of these six conserved cysteine residues (all but cysteine 4) had d_N < d_S values, where the difference was significant in multiple HyPhy modules (Table 3). The fourth cysteine has the same codon (TGC) for all sequences in this alignment (as well as in the Fig. 1a alignment excepting the Microctonus and one C. elegans ITP sequence), explaining why this codon was not detected as negatively selected. This high degree of conservation of cysteine codons is similar to the pattern observed in conotoxins, where there are high rates of nonsynonymous substitutions in most codons, but a striking uniformity of cysteine codons across sequences (Conticello et al., 2001; Duda and Palumbi, 1999).

4.3 Function of latrodectins in black widow venom

Previous biochemical studies of black widow spider venom indicated latrodectins constitute a major fraction of the venom, with latrodectin 1 being in approximately equal quantities with α-latrotoxin (Grishin et al., 1993; Kiyatkin et al., 1992; Volkova et al., 1995). Latrodectins only comprised 3-4% of all ESTs we cloned from Latrodectus and Steatoda venom gland cDNA libraries, but they were among the few transcript types represented by multiple copies in the libraries, consistent with their relatively high expression in venom glands. Given the abundance of latrodectins in black widow spider venom, and evidence of paralogs that experienced negative and positive selection, it is likely that latrodectins plays some key role in prey acquisition. However, past functional studies indicated that purified latrodectins are not toxic to mammals or insects (Kiyatkin et al., 1995; Volkova et al., 1995). For example, purified latrodectin 1 and 2 from L. tredecimguttatus is non-toxic to cockroaches (Periplaneta americana) when injected at doses of up to 80 μg/g, having not caused lethality or paralysis (Volkova et al., 1995). Purified L. tredecimguttatus latrodectins were also not toxic to mice at concentrations up to 2.3 mg/kg intravenously, and 0.8 mg/kg intracerebroventricularly (Volkova et al., 1995). Interestingly, TaITX-1-3, proposed latrodectin homologs from Tegenaria agrestis (hobo spider) venom, are characterized as highly potent insect neurotoxins that kill via paralysis and cause elevated firing of neurons from the central nervous system, although the mechanism of insecticidal activity is unclear (Johnson et al., 1998). This suggests that spider venom peptides derived from the CHH/ITP neuropeptide family exhibit substantial functional diversity.

Latrodectins typically co-purify with latrotoxins and are biochemically challenging to separate (Grishin et al., 1993; Kiyatkin et al., 1992; Volkova et al., 1995). Latrodectin 1 was first isolated in an attempt to isolate α-latrotoxin (Kiyatkin et al., 1992), and latrodectin 2 was isolated with latroinsectotoxin fractions (Volkova et al., 1995). This suggested that latrodectins interact with latrotoxins as heteromeric complexes (Kiyatkin et al., 1992). Subsequent work showing that recombinant α-latrotoxin can recapitulate the effects of whole venom in causing neurotransmitter release in the absence of latrodectin 1 (Volynski et al., 1999), casts some doubt on whether α-latrotoxin and latrodectin 1 function together. Nevertheless, Grishin et al. (1993) used latrodectin 1-specific antibodies to show that inhibition of latrodectin 1 in the presence of α-latrotoxin resulted in a decrease in calcium ion entry in synaptosomes relative to α-latrotoxin alone, but latrodectin 1 inhibition also resulted in an increase in neurotransmitter release in a separate experiment. Despite these somewhat mixed results, Grishin et al. (1993) suggested that latrodectins work to enhance α-latrotoxin function by promoting calcium influx and neurotransmitter release. The derivation of latrodectins from ion transport peptides (ITP) suggests that their function in black widow venom may involve modulating the transport of calcium ions around or near nerve cells, potentially to enhance the toxicity of α-latrotoxin. For example, locust Ion Transport Peptide (ITP) functions to control the movement of chloride anions across cellular membranes of the hindgut (Audsley et al., 1992). Moreover, latrotoxins and latrodectins could form a complex that in vivo negatively affects neurons by offsetting ion balance near different channel types, which may or may not contribute to neurotransmitter release. While the evolutionary analyses we present reinforces the functional importance of latrodectins in black widow spider venom, more extensive neurophysiological and biochemical studies of their structure and activities in relation to the better understood latrotoxins are clearly needed to determine their function in whole venom.

4.4 Conclusions

Using phylogenetic analyses and evidence from gene structure, we show that latrodectin peptides from black widow spider venom are derived from the ecdysozan superfamily of neuropeptides containing Crustacean Hyperglycemic Hormones (CHH) and Ion Transport Peptides (ITP). Evidence of positive and negative selection operating on latrodectin sequences supports an important functional role of these peptides in black widow venom. Our phylogenetic evidence also suggests the deployment of CHH/ITP homologs for venom expression in additional arthropod taxa, consistent with the independent recruitment of homologous genes for venom expression. These findings should be investigated as comprehensive transcriptomic and genomic resources are developed for neglected venomous taxa.

Highlights.

• Latrodectin venom peptides evolved from the CHH/ITP neuropeptide hormone family.

• We sampled latrodectin cDNAs and genes from black widows and related species.

• CHH/ITP genes were recruited for venom expression multiple times in arthropods.

• Latrodectin and CHH/ITP genes share a conserved intron position.

• Black widow latrodectin genes have experienced positive and negative selection.

Abbreviations

aa

amino acid(s)

AIC

Akaike Information Criterion

α-latrotoxin LMWPs

α-latrotoxin associated Low Molecular Weight Proteins

BEB

Bayes Empirical Bayes

bp

base pair(s)

cDNA

DNA complementary to RNA

CHH

Crustacean Hyperglycemic Hormone

CPP

Clade Posterior Probability

d_N

nonsynonymous substitutions per nonsynonymous site

ds

double strand(ed)

d_S

synonymous substitutions per synonymous site

EST

Expressed Sequence Tag(s)

FEL

Fixed Effects Likelihood

FUBAR

Fast Unbiased Bayesian AppRoximation

GTR+I+G

Generalized Time Reversible plus Gamma plus Invariant model

ICK

Inhibitor Cystine Knot

ITP

Ion Transport Peptide

LRT

likelihood ratio test

MEME

Mixed Effects Model of Episodic Diversifying Selection

ML

Maximum Likelihood

NCBI

National Center for Biotechnology Information

RTA

Random Taxon Additions

RTA clade

Retrolateral Tibial Apophysis clade

SLAC

Single-Likelihood Ancestor Counting

TPM3uf +I+G

Three Parameter with unequal base frequencies plus Gamma plus Invariant model