In some reptiles and other animals, protein recruiting to specialized glands and their diversification throughout evolution resulted in venoms, which represent a key evolutionary innovation used by these animals for defense, competitor deterrence or predation 1 . When inoculated into a prey/predator, these proteins target important physiological pathways and tissue types. Toxins impairing prey motility have been particularly selected during this process being the neurotoxins available in almost all taxons of venomous animals. On the other hand, the distribution of toxins by the bloodstream makes the hemostatic system also an important venom target. Despite the functional versatility of venom proteins, only a few protein families are frequent in venom composition and this aspect is compensated by a high degree of sequence variability, within the superfamily, generated by accelerated evolution 2 .

In viper snake venoms, the most abundant protein families are serine proteinases, metalloproteases, L-amino acid oxidases, phospholipases A₂ (PLA₂s), disintegrins, C-type lectins, myotoxins, and CRISP toxins 3 , represented by a number of distinct copies with variable structural motifs and functional activities. To explain the generation of such diversity, evolutionary relationships of venom PLA2s 4 5 , C-type lectin-like proteins 6 , serine proteinase inhibitors 7 , and metalloproteinases 8 described that these different sequences evolved by gene duplications followed by positive selection of non-synonymous mutations accumulated in the genomes under strong positive adaptive selection. High levels of variation in venom proteins would confer an advantage to adapt to a variety of different prey 9 10 . Besides the diversity generated during evolution, sequence variability may further increase during post-translational modifications of nascent toxins 11 and by differential expression of toxin groups according to snake geographical distribution 12 , sex 13 or ontogenesis 14 15 . For example, in Costa Rica, toxin profiles of Bothrops asper venom from adult snakes are characterized by the predominance of PI-SVMPs and K49-PLA2s 16 whereas in Brazil, venom of adult Bothrops atrox present the paedomorphic phenotype with predominance of PIII-SVMPs and D49-PLA2s 17 . The paedomorphic phenotype is prevalent in the most important species of Bothrops snakes in Brazil, with predominance of P-III SVMPs in venom proteomes or transcriptomes 18 19 20 .

SVMPs are encompassed in the M12 subfamily of metalloproteinases (reprolysins). They are composed of an independent proteolytic domain or of combination between the proteolytic domain and additional adhesive/anti-adhesive domains, as disintegrin or disintegrin-like/cysteine-rich domains, preceded by a N-terminal well conserved pro-domain. Following different steps of post-translational modifications, distinct types of SVMPs may be found in viper venoms, being originally derived from distinct precursor molecules. SVMPs were formerly classified as four distinct classes of nucleotidic sequences (N-classification) that after translation would give rise to four classes of mature proteins (P-classification) with different domain organization. Precursors belonging to class N-I (pre-, pro- and catalytic domains), N-II (pre-, pro-, catalytic and disintegrin domains), N-III (pre-, pro-, catalytic, disintegrin-like and cysteine-rich domains) and N-IV (pre-, pro-, catalytic, disintegrin-like, cysteine-rich and lectin-like domains) would allow biosyntheses of P-I to P-IV SVMPs that include the same domains present on precursors, but with removal of pre- and pro-domains. According to this classification, catalytic domains of P-II, P-III and P-IV SVMPs could also be proteolytically removed to generate P-I SVMPs 21 . However, N-IV precursors have never been found in gland transcriptomes and independent catalytic domains of P-III or P-IV SVMPs have never been isolated as a P-I SVMP in snake venoms. Therefore, classification of SVMPs was reviewed several times and what is currently accepted is the same terminology for N- or P- classes, now referred only as P-classes, and inclusion of the N/P-IV class as a post-translational modification of a class P-III SVMP, presently classified as P-IIId 11 . It is now accepted 11 different subclasses of SVMPs with the following distribution: one type in class P-I, 6 in class P-II and 4 in class P-III, each generating different possibilities of mature proteins depending on dimerization or proteolytic release of independent disintegrin or disintegrin-like domains 11 . The presence and structural features of the domains characterize the function and toxicity of SVMPs. In this regard, SVMPs play important roles in hemostatic disorders and local tissue damage that follows snakebite showing diverse effects both on soluble factors and cellular components. The action of SVMPs involves catalytic and adhesive/anti-adhesive properties, as well as direct cellular activation and/or the release of endogenous bioactive components. They hydrolyze extracellular matrix, plasma and cell-surface proteins, interfere with blood coagulation and fibrinolysis, inhibit platelet aggregation, and show a diversity of activities towards inflammatory and endothelial cells compromising the integrity of capillary vessels 22 .

To understand the structural complexity and functional diversity of SVMPs, some studies attempted to explain the recruitment and evolution of this protein family within venom glands of viper snakes. In early studies, we suggested a common ancestry of mammalian matrix metalloproteinases (MMPs) with a P-III-type of SVMP that underwent divergence after gene duplications generating P-II and P-I classes scaffolds 8 . Further studies attempted to elucidate the evolutionary history of SVMPs and one important aspect that has been speculated is that loss of domains in the P-III ancestor gene could be attributed to the neofunctionalization of the disintegrin domain in some of the duplicated P-III genes 23 24 . Recently, Casewell et al. 25 proposed that the evolution of SVMPs is repeatedly punctuated by domain loss to form multiple P-I and P-II SVMP structures. However, it remains untested if the gain of domain structures through mechanisms as exon shuffling or pre- or post-transcriptional recombination could also play a role in the diversification of SVMP structures.

In this study, we have unveiled the diversity of SVMPs expressed by a specimen of Bothrops neuwiedi snake by cloning, sequencing and analyzing the SVMP cDNAs transcribed in a single venom gland. A great diversity in the SVMP repertoire was confirmed. Moreover, evidences emerged for recombination between genes of SVMP sequences from different classes and also between their primary mRNA transcripts, enlarging the possibilities of mechanisms involved in generation of diversity of this class of toxins and defining new perspectives in the classification and function prediction of this group of toxins.

Recently, transcriptomics and proteomics approaches have been extensively used to analyze the global composition of venoms from viper snakes. These studies provide a comprehensive knowledge of primary structure data based on sequences available in databanks. They are very important for an accurate description of the protein families expressed allowing evolutionary inferences about the generation of venom diversity and the toxin arsenal present in snake venoms 27 28 . In most of the referred studies, SVMPs appear as the major component in venoms of viper snakes and representatives of P-I, P-II and P-III classes were detected in all of them. However, these approaches are not the most appropriate for inferences of functional diversity since sequence characterization is usually based on molecular fragments of varying sizes that may omit structural rearrangements or motifs involved in important functions of a particular toxin. In this study, thirty-one cDNAs coding for mature SVMPs were completely sequenced from a gland of one specimen of B. neuwiedi snake. Internal alignments and phylogenetic inferences with databank SVMPs revealed new aspects of the evolution of functional diversity of this protein family not evidenced before by high throughput approaches.

Sequences of SVMPs from classes P-I, P-II and P-III were amplified from B. neuwiedi venom gland with particular abundance of cDNA sequences from class P-II SVMPs (BnMP-II). Representatives of each class were not identical, with at least two distinct copies of P-I cDNAs (with 98% identity) and three of P-III (with % identity values above 88). In opposition, great variability was observed amongst class P-II sequences with six distinct clones presenting a greater number of nucleotide substitution and divergence (identity from 73%). High sequence variability is not surprising since, considering that toxins evolved by accelerated evolution, a number of non-synonymous substitutions is predicted in coding regions driving to diversity of venom proteins, including SVMPs. Indeed, nucleotide substitutions were apparently required for the emergence of new functions as disclaimed for disintegrins 24 . Moreover, toxin genes contain in their exons a greater number of unstable triplets that were subjected to a faster point-mutation rate than constitutive genes 29 . In this regard, we suggest that point mutations are occurring in SVMPs with a higher frequency in representatives of class P-II, reflecting a higher plasticity in this class of genes that reflects in the generation of new functions in the venoms.

Besides the diversity of class P-II SVMPs generated by nucleotide substitutions, evidences of recombination between P-II disintegrin domain with catalytic domains of P-I or P-III SVMPs were also noted. When B. neuwiedi sequences were clustered by cladistic analysis, three distinct types of P-II sequences were detected with distinct branching: BnMP-IIa grouped with P-I SVMPs, BnMP-IIx with P-III, and BnMP-IIb in a third group, composed only of P-II SVMPs. This evidence suggests that recombination between genes encoding SVMPs might have occurred after the emergence of the primary gene copies coding for each scaffold. In their recent publication, Casewell et al. 25 suggest that different scaffolds of SVMPs were generated by minimization of genes with domain loss, which have occurred several times during evolution of this protein family. The loss of cysteine rich domain would have occurred once to generate P-II SVMPs and the loss of disintegrin domains would have occurred at least 8 times in P-II to generate P-I SVMPs. Indeed, this is a very plausible hypothesis already supported by other reports in the literature 8 24 . The data presented in this study is not in disagreement with this theory. However, we add an alternative mechanism to explain the diversity of precursors coding for SVMPs. The topology of the trees shown in this study indicates recombination of P-II genes with members from P-I and P-III classes both at DNA and RNA levels after the emergence of the three different scaffolds. The hypothetical recombination between P-I and P-II SVMPs might occur at the post-transcriptional level, supported by the striking identity (100%) between the 5'-terminal 365 bp fragment of BnMP-I and BnMP-IIa cDNA sequences that apparently were alternatively joined with two distinct fragments at 5', which diverged by nucleotide substitutions and the inclusion in only one of the fragments the region coding for the disintegrin domain. According to the previous papers 8 24 25 , P-I SVMPs were derived from a PII ancestor and this would imply high similarity between such precursors. However, considering that genes coding for SVMPs are subjected to accelerated evolution and that most of substitutions accumulate in coding regions, it is very unlike to expect 100% identity in cDNA regions of genes that diverged previously, even though the time of divergence between P-I and P-II SVMPs is not reported in previous papers 8 24 25 . In this regard, post-transcriptional recombination is an appealing hypothesis to explain, together with genetic events, the generation of class P-I SVMPs in different events of SVMPs evolutionary history, as reported previously 25 . Corroborating this hypothesis we found complementary regions in B. neuwiedi cDNA sequences that could be involved in loop assembly in the RNAs enabling them for alternative trans-splicing leading to the formation of "hybrid" mRNAs. Evidences for alternative splicing have been shown in Bungarus fasciatus acetylcholinesterase gene generating different proteins in liver or venom gland 30 . This was also suggested by Siigur and colleagues 31 , when studying the generation of diversity of serineproteinases from Vipera lebetiba venom and, more recently, Zeng et al. 32 provided evidences of trans-splicing for generation of toxin diversity in the venom of Buthus martensii scorpion. Alternative splicing may explain the intriguing distribution of class P-I SVMPs in previous phylogenetic studies of SVMPs, also grouped with class P-IIa SVMPs in different branches of their phylogenetic tree 25 .

The other evidence of recombination between different classes of SVMPs was deduced from the phylogenetic inferences of independent domains. Several studies report the phylogenetic relationships of SVMPs isolated from closely-related species, or of specific classes or domains 8 23 24 33 34 35 36 . Usually, these studies deal with a large number of sequences with different degrees of similarity. In this study, the hypothesis of interclass recombination arose from experimental data, during sequence alignments. Thus, the analyses were carried out using a different approach driven to identify such recombination in other snake species. For this purpose, we selected by megablast 37 only the most similar sequences (>85% identity) of independent domains from each BnMP sequences and carried out the phylogenetic analysis with this selected group of toxins. This approach may not represent the evolutionary history of the protein family, but allowed the characterization of closest relationships between catalytic domains from different classes of SVMPs, with statistically supported clusters enclosing P-I and P-IIa (99%) or including P-IIx and P-III (100%) catalytic domains. Considering the disintegrin domain, P-IIa, P-IIb and P-IIx sequences were monophyletic (94%). We understand that this is strong evidence that the multiple domain structure observed in SVMPs has undergone different evolutionary history, and the protein variability of this protein family may be dependent on recombination at different points of the evolution of this protein family. In the case of BnMP-IIx and P-III SVMPs, recombination events might occur at the genomic level with the assembly of P-II disintegrin domains with catalytic regions from genes of non-hemorrhagic SVMPs class P-III. The possibility that genomic recombination of independent domains occurs by exon shuffling cannot be ruled out since this is an accepted mechanism for evolution of modular molecules, particularly plasma and extracellular matrix proteins 38 39 . In venoms, segment switches in exons (ASSETS) have been demonstrated and correlated to targeting diversification 40 . Moreover, it has also been reported by different authors the occurrence of SVMP structures that might have been assembled by P-III catalytic domain with P-IIa disintegrin domain 41 or by PIIa catalytic domains with P-III disintegrin-like domains lacking the cysteine-rich domain 42 . These evidences supporting recombination between P-III catalytic and P-II disintegrin domains indicate that the loss of cysteine rich domain may be a more frequent mechanism and may have occurred in different events during evolution of SVMPs. Unfortunately, these mechanisms of recombination are still speculative since, up to the moment, genomic sequences coding for SVMPs were not completely disclosed and the exon/intron distribution at catalytic domain is still unknown.

Regardless of the mechanisms used for generation of new genes/mRNAs encoding class P-II SVMPs, this is certainly the class under highest adaptative pressure and positive selection. Soto et al. 36 showed that accumulation of mutations is less frequent in class P-III SVMPs than in class P-II. In this study, we noted in a single snake three independent P-II structures, coding for distinct proteins. BnMP-IIa sequence shares catalytic domain with P-I SVMPs and representatives enclosed in this cluster generate, by processing, the P-I SVMPs (only catalytic domains) and the classical RGD-disintegrins; BnMP-IIb sequences grouped with other class P-IIb SVMPs corresponding to the non-proteolytic processed P-II SVMPs; BnMP-IIx group corresponds to a new subclass of P-II SVMP precursor which predicts proteins with catalytic domain of non-hemorrhagic P-III SVMPs and a disintegrin domain with a higher number of cysteines, an extra pair at C-termini that may be involved in a new disulfide bond reported here for the first time. This domain association has not been reported yet and the presence of these molecules in venoms in the unprocessed form is still unknown and the possibility that this precursor is a pseudogene is not ruled out up to the moment. This distribution is in agreement to the gene tree reported by Casewell et al. 25 with a few differences: the authors include a fourth group with dimeric P-IId SVMP and did not evidence the P-IIx SVMPs. It is important to note the large number of sequences enclosed on their P-I/P-IIa group. In our case, BnMP-IIa was the most abundant cDNA in the gland of B. neuwiedi, but in the pool of B. neuwiedi venom, the most abundant proteins are two isoforms of a class P-I SVMP, BnP1 and BnP2, which were partially sequenced revealing approximately 90% identity with either BnMP-I or BnMP-IIa cDNAs 43 . Taken together, the great number of P-I/P-IIa SVMPs described and their high expression levels in venom glands point to the advantage of this form of precursor since it generates both P-I SVMPs and free disintegrins from a single precursor. The possible recombination between P-I and P-II precursors shown in this study is of great relevance for the evolution of SVMPs and the generation of the functionally active representatives of this family of proteins.

A great diversity was observed in cDNAs encoding SVMPs in a single venom gland of a B. neuwiedi specimen. The most complex group was the class P-II represented by three independent subclasses with evidences that they may be arisen by recent recombination either at genomic level or post-transcription between different classes of SVMPs. This includes a new subclass of P-II SVMP precursor, P-IIx, composed of a combination of P-III catalytic domain and a RGD disintegrin domain with extra cysteines at C-termini. These evidences add new possibilities of SVMPs scaffolds by the inclusion of a P-IIx type of SVMP. In regard to biosynthesis of the P-I group, at least in our set of precursors, P-I proteins appeared to be synthesized from both P-I and P-II precursors, and it is very likely that P-I precursors were derived from P-II post-transcriptionally. Moreover, our data enlarge the possibilities of mechanisms involved in generation of diversity of this class of toxins. However, a complete understanding of the genesis of venom diversity and definite evidences of recombination are still dependent on the resolution of the genome of a viper snake.

Conceived and designed the experiments: GSM, RHV, AMMS. Performed the experiments: KFG, MSF, MCC, PBC, JAPJ. Analyzed the data: GSM, PBC, RHV, AMMS. Wrote the paper: GSM, PBC, JAPJ, RHV, AMMS. All authors read and approved the final manuscript.