Venom is rare across mammals. The most famous example of venom in mammals is the male platypus (Ornithorhynchus anatinus), which the animal delivers from a spur on the calcaneus. Since the tarsal spur is a sexually dimorphic trait, it likely functions for mate defence in the platypus (Ligabue-Braun et al., 2012). The slow loris (genus Nycticebus) secretes a protein from the brachial gland near its elbow, which, when mixed with saliva and injected via a bite, produces anaphylaxis in a victim. Lorises apply their brachial secretions to the fur around their head, and appear to use them primarily as warning signals to conspecifics (Hagey et al., 2007). The toxin seems to be a secondary function of the exudate, and presumably affects only susceptible species like humans (Hagey et al., 2007).

In extant eutherian mammals, the only order containing members with venomous saliva is the Eulipotyphla (hedgehogs, shrews, moles & solenodons) – this order contains about 452 species (Wilson and Reeder, 2005). Within Eulipotyphla, only four species have been shown to be venomous: the Hispaniolan solenodon (Solenodon paradoxus) [the Cuban solenodon (S. cubanus) is possibly also venomous, although it has not been tested] and three of 376 species of shrews (Soricidae) (Blarina brevicauda, Neomys fodiens and N. anomalus) (Dufton, 1992 and Pucek, 1968). There is some evidence that other species of Blarina (B. carolinensis and B. hylophaga) and the Canarian shrew (Crocidura canariensis) (Lopez Jurado and Mateo, 1996) may be venomous; prey bitten by these animals reacts in a similar fashion to prey bitten by known venomous animals, but their saliva has not been tested experimentally (Lopez Jurado and Mateo, 1996). Many other species are as yet untested (including the American shrew, Sorex cinerus, and the European mole, Talpa europaea) (Dufton, 1992 and Ligabue-Braun et al., 2012). Nussbaum and Maser (1969) observed the water shrew Sorex palustris immobilize a large vertebrate prey by grasping it by the head and interpreted this as evidence of venom; however, Pearson (1956) reported no significant toxicity in S. palustris saliva.

However, a number of species of eulipotyphlans and rodents have been tested and found to lack venom. These include: the European hedgehog (Erinaceus europaeus) (Mebs, 1999), the common mole (Scalopus aquaticus), the white-footed mouse (Peromyscus leucopus), meadow mouse (Microtus pennsylvanicus) (Pearson, 1942), a number of long-tailed shrews (Sorex fumeus, S. pacificus, S. palustris, S. sinuosus, S. trowbridgii, S. personatus and S. vagrans), the least shrew (Cryptotis parva), the shrew-mole (Neurotrichus gibbsi) (Pearson, 1950 and Pearson, 1956), and the greater white toothed shrew (Crocidura russulla) (Bernard, 1960). These observations suggest that, while not all species have been evaluated, venom is relatively rare across Eulipotyphla.

Of the 376 species of living shrews, only three are known to be venomous (0.8%). Even if we include species that are reasonably likely to be venomous – B. carolinensis, B. hylophaga, Crocidura canariensis – the proportion of venomous shrew species only increases to 1.6%. If we look at the proportion of all eulipotyphlans that are or may be venomous (including the solenodon), there are four known venomous species out of 452 (or eight if we extend it to the possibly venomous species, including S. cubanus). This means that 0.9% (at most 1.8%) of eulipotyphlans are venomous. If we assume that the proportion of modern venomous taxa accurately represents the distribution of venom in extinct species (and there is no good reason to assume that venom was more common in the past), we would predict that a similarly tiny fraction (less than 2%) of extinct species of shrew should be venomous.

Assessing the phylogenetic relationships among venomous shrews might shed light on how the trait evolved. Is it, as some have suggested (e.g. Hurum et al., 2006), a primitive trait that was lost in non-venomous lineages? Or is it an evolutionary novelty in a few eulipotyphlans? If it is a novel trait, how many times did it evolve? To address these questions, I mapped the incidence of venom onto the most recent estimates of eulipotyphlan phylogeny (Fig. 1). Dubey et al., 2007 and Dubey et al., 2008 used both mtDNA and nuclear DNA to reconstruct relationships among shrews (Soricidae). He et al. (2010) extended this data set by incorporating additional sequences and several new taxa. As their results are congruent with Dubey et al., 2007 and Dubey et al., 2008 tree topology, I constructed a time-calibrated supertree combining the data from all three studies, plus estimates of the placement of basal eulipotyphlans Erinaceae and Solenodontidae (Douady and Douzery, 2009 and Roca et al., 2004) (Fig. 1). This topology is consistent with Ohdachi et al.’s (1997) relatively unresolved tree, which also found the monophyletic groups (Cryptotis, Blarina), (Episoriculus, Nectogale, Neomys) and (Crocidura, Suncus). An alternate topology by Willows-Munro and Matthee (2011) only included 664 base pairs from the cytochrome b gene, and resulted in a different tree topology. Although the Willows-Munro and Matthee (2011) sample was more taxonomically diverse, the larger datasets of Dubey et al., 2007 and Dubey et al., 2008 and He et al. (2010) include the cytochrome b gene as well as additional mtDNA and nuclear DNA, and is the topology I use here. There is some disagreement over the position of moles (Talpa) relative to the Soricomorph clade (Solenodon, Sorex and Erinaceus). In a recent combined morphological and molecular analysis of placental mammals, O’Leary et al. (2013) found Talpa to be basal to the Soricomorphs. This would not significantly change the tree topology, as Solenodon is still basal to a clade of (Sorex, Erinaceus).

This phylogeny demonstrates that Neomys, Solenodon and Blarina brevicauda are phylogenetically distant; the Solenodon lineage split from the other eulipotyphlans around 80.5 mya (Douady and Douzery, 2009) and the ancestors of Neomys and Blarina split around 16.5 mya (Dubey et al., 2007) (Fig. 1). If the ancestral condition in Eulipotyphla is to possess venom, with the primitive retention of the trait in Blarina and Neomys and the loss of venom in other lineages, it would require a minimum of nine convergent losses of venom. However, if venom evolved convergently, it would only require three unique acquisitions. Thus, the most parsimonious explanation of these data is that venom evolved independently in Solenodon, Neomys and Blarina from non-venomous ancestors.

The observation that venom evolved multiple times in the eulipotyphlan lineage might suggest that this small mammal group is exapted (sensu Gould and Vrba, 1982) to acquire the ability to secrete venom proteins. The synthesis of venom is complex, so maybe there is a chemical present in the salivary glands of all soricids or eulipotyphlans that allows these taxa to independently acquire venom more easily than in other species. However, the acquisition of venom is rare, occurring in less than 1% of eulipotyphlans, so either it is not particularly adaptive or it is constrained by other factors.

Recently, several fossil eulipotyphlans have been described in which venom has been inferred on the basis of grooved teeth (Cuenca-Bescós and Rofes, 2007, Peigné et al., 2009, Rofes and Cuenca-Bescós, 2009 and Turvey, 2010). An inference of venom in fossil mammals can be made based on the following criteria: •

if grooved teeth are used as the basis of an inference of venom, the morphology of the grooves should be similar to other venomous mammals with grooved teeth (i.e. only solenodons);

However, even if a single fossil species can be shown to be venomous based on these criteria, it does not tell us much about the overall significance of venom in mammal evolution.

To apply the first criterion to soricids, most shrews lack grooved teeth, including the venomous shrews (Blarina brevicauda and Neomys). Venomous shrews salivate copiously (Churchfield, 1990 and Pucek, 1959), and envenomate their prey by delivering multiple rapid but shallow bites to prey, aiming for the head and neck (Bernard, 1960, Churchfield, 1990, Eadie, 1949, Haberl, 2002 and Tomasi, 1978). Some of the lower incisors of Blarina and Neomys show a slight concavity, as do the teeth of a number of other, non-venomous soricids (Pournelle, 1968); however, this very slight guttering is unlikely to function as a means of delivering venom. Venom entrance into a wound appears to be transferred somewhat haphazardly, unlike the needle-like delivery system of snakes, and it is not clear that the slight concavity in the lower teeth serves any function in delivering venom. Many non-venomous shrews also have slightly concave lower incisors (Churchfield, 1990; Pournelle, 1968).

Because no extant shrew possesses both grooved teeth and venomous saliva, grooves on the teeth of fossil shrews are unlikely to be related to venom delivery. Solenodons have venom delivery teeth – the lower second incisors form tubes or channels that are suggested to act like hypodermic needles (Dufton, 1992 and McDowell, 1958) – as well as venomous saliva. Only Solenodon paradoxus has been shown to possess venom, although S. cubanus has not been tested, and two other species recently went extinct. Therefore, fossil solenodon relatives with channeled lower incisors similar to S. paradoxus are likely to share the trait through homology, and could be inferred to have also had venom.

Cuenca-Bescós and Rofes (2007) and Rofes and Cuenca-Bescós (2009) described two species of large shrew from the Pleistocene of Sierra de Atapuerca, Spain – Beremendia fissidens and Dolinasorex glyphodon. On the basis of a groove on the mesial surface of the lower incisors, they infer that these species were likely to have been venomous. Folinsbee et al. (2007) made the assessment that the inference of venom delivery in Beremendia and Dolinasorex was better supported than for the pantolestid Bisonalveus browni (Fox and Scott, 2005), because the former are members of the Eulipotyphla, which contains venomous members. However, this statement bears closer examination – Eulipotyphla is a large clade containing around 452 species (Douady and Douzery, 2009), 99% of which are not venomous. Since no known living shrews possess both grooved teeth and venom, inferring venom on the basis of a mesial groove on the incisors cannot be supported. In fact, the morphology of the groove in the lower left incisors of these two species looks very similar to the grooved lower second incisor of the eastern mole (Scalophus aquaticus) (Fig. 2), which has been shown to lack venom (Pearson, 1942).

When we consider Eulipotyphlan phylogeny (Fig. 1), it becomes apparent that venom is a rare trait that likely evolved convergently at least three times. Given the evolutionary rarity of the trait at present, it is unlikely to have been widely distributed in the past. Rofes and Cuenca-Bescós (2009) perform a phylogenetic analysis of morphological characters including both extinct and extant species. Their tree topology is quite different from that of Dubey et al., 2007 and Dubey et al., 2008 (Fig. 1); however, like Dubey et al., 2007 and Dubey et al., 2008 and He et al. (2010), they reconstructed a monophyletic group that includes [Soriculus (Chimarrogale, Nectogale)]. Rofes and Cuenca-Bescós (2009) determined that the extinct genera Dolinasorex and Beremendia fall within this monophyletic group. An inference of venom might be warranted, although not on the basis of grooved teeth, if Dolinasorex and Beremendia were the sister taxa of Neomys or Blarina. However, they are only distantly related to known venomous shrews, and therefore there is no support for the hypothesis that they inherited venom from a venomous ancestor.

Furió et al. (2010) suggested that Beremendia used venom to hoard small coleopteran and gastropod prey, in the same way that living venomous shrews do, to contend with the unpredictable Pleistocene environmental conditions. They claim that in living shrews “[t]here is no doubt that the lower incisor is always the main element involved in this venomous injection, and that the groove acts as a channel to conduct the venom from the salivary glands to the anterior part of the lower incisor. Nevertheless, although the grooves do exist in most of the species (Fig. 5), they are usually difficult to discern due to their shape and/or smaller dimensions.” (2010: 933). Small, shallow concavities in the lower incisors may indeed exist in Sorex and Crocidura (as shown in Furió et al.’s fig. 5); however, neither Sorex nor Crocidura possess venomous saliva (Pearson, 1942, Pearson, 1950 and Pearson, 1956). Both Blarina and Neomys have a slightly concave lingual face of the lower first incisor, which has been suggested to function to transmit venom (Pournelle, 1968), but this has not been tested. In fact, the shallow concavity on the lingual face of the lower central incisors in Blarina and Neomys could only possibly function as a location for venom to pool, but it is too shallow, open and small to “act as a channel to conduct venom”. An alternative hypothesis is that the groove serves a mechanical function to resist bending.

The shallow concavities in the extant venomous shrews are very different in morphology from the deeper grooves in Beremendia, Dolinasorex and Scalopus. Both Beremendia and Dolinasorex were large, estimated to be about 40–45 g and 55–60 g, respectively (Cuenca-Bescós and Rofes, 2007 and Rofes and Cuenca-Bescós, 2009). Moles are also larger than most shrews, with a body mass of about 65 to 120 g for the European mole and 40 to 140 g for the Eastern American mole (Nowak, 1999). The difference in the size of the concavity or groove in the lower incisors between small (e.g. Neomys, Blarina, Sorex and Crocidura, all of which are less than 30 g) and large eulipotyphlyans (e.g. Scalopus, Talpa, Beremendia and Dolinasorex) may simply be a result of allometric scaling. Small shrews tend to have slightly concave lower incisors, and larger animals with larger teeth have correspondingly larger grooves. Because the presence of a groove in the lower teeth is not restricted to venomous taxa, we cannot automatically assume that it functions to deliver venom.

Peigné et al. (2009) describe a large soricoid, Siamosorex debonisi, from the Oligocene of Thailand. They suggested on the basis of grooved second incisors and its phylogenetic position as a soricid that S. debonisi might have had venomous saliva. Similar to Beremendia and Dolinasorex, Siamosorex is relatively large in size, and therefore a simple allometric relationship can explain the presence of a slightly deeper groove in the teeth. Peigné et al. (2009) qualified their assessment, stating that it is almost equally as likely that Siamosorex lacked venom.

Although these extinct shrews have some unique traits, they do not meet either criterion for inferring venom: they do not have grooved teeth that are similar to grooved teeth in venomous animals, nor are they closely phylogenetically related to known venomous taxa. Venom in fossil shrews cannot be inferred from the presence of deeply grooved teeth because no extant shrew possesses both deeply grooved teeth and venom.

Turvey (2010) described the unique dental features in the recently extinct genus of West Indian eulipotyphlan, Neophontes, which he inferred to be related to venom delivery. He described two deep grooves the upper canines, which look similar to the deep grooves in the lower incisors of solenodons. As Turvey stated, this would require the convergent evolution of toxic saliva from modified parotid glands, rather than from modified submaxillary glands as in all other orally delivered mammalian venom (Dufton, 1992 and Pearson, 1950). The phylogenetic placement of Nesophontes is uncertain, but it appears to be either the sister-taxon of Solenodon, or to fall deeper within the Eulipotyphla clade as the sister to the soricids (Asher et al., 2005, Roca et al., 2004 and Turvey, 2010). However, as there is no known mammal with modified parotid glands that secretes venom, this inference must remain tentative. Interestingly, since these animals are recently extinct, there is the possibility that DNA analysis might reveal genes for venom production.

Unfortunately, modern venomous eulipotyphlans, with the exception of solenodons, give no evidence of their venomous capabilities from their skulls and teeth alone. If Blarina or Neomys were found in the fossil record, we would never infer that they possessed salivary venom. Conversely, the deeply grooved lower second incisors of living eastern moles (Fig. 2), as well as the canines of baboons and bats (Folinsbee et al., 2007 and Orr et al., 2007) could suggest to paleontologists that venom is more widely distributed than it is. This starkly illustrates the difficulty in inferring soft tissue-based traits from the fossil record.

A bias toward reporting purportedly “venomous” mammals may result in their overrepresentation in the published literature. Shrews are small, speciose and exist in restricted endemic home ranges. Many species of fossil shrews are unlikely to be thoroughly described in the literature, unless there is a novel aspect about their biology.

The hypothesis that the earliest mammals were venomous has been proposed previously, based on the notion that modern venomous mammals are found in “primitive” clades (i.e. Dufton, 1992, Fox and Scott, 2005 and Hurum et al., 2006). Dufton (1992) suggested that all early placental insectivores may have been venomous, with only a few species retaining the trait. He proposed that other eulipotyphlans lost the capacity to produce venom because they were too successful as predators and outcompeted their vertebrate prey. This, he argued, resulted in a shift to an insectivorous diet in early eulipotyphlans, and a subsequent loss of the need for venom in all but a few species. This explanation is unlikely for several reasons: first, venomous shrews (Blarina and Neomys) are engaged in stable predator-prey webs with rodents (e.g. Microtus) without causing a precipitous decline in these vertebrate prey species (Barbehenn, 1958, Bernard, 1960, Eadie, 1948 and Eadie, 1952); natural history observations do not bear out the prediction that venomous shrews are capable of decimating rodent populations, nor has this phenomenon been observed in nature in other species. Second, because venom is so rare and widely dispersed across eulipotyphlans (Fig. 1), this hypothesis would require the homoplastic loss of venom in at least nine lineages. Finally, venom seems to be incredibly effective for preying on invertebrates, which make up the vast majority of the diet in Blarina and Neomys (Pournelle, 1968). Therefore, it does not make sense that venom would have fallen into disuse across the majority of eulipotyphlan lineages merely because they switched to invertebrate prey; Blarina and Neomys use their venom to great advantage on invertebrates as well as vertebrates. The rarity of venom, and the fact that it is found in relatively derived eulipotyphlans, suggests that it evolved convergently rather than being a primitive retention.

But how could such complex chemical neurotoxins evolve repeatedly in distantly related species? Reptile venom evolution has been the subject of much more intensive study than venom in mammals, and may provide some insight into how salivary venoms convergently evolve. Salivary enzymes, including lysozymes and amylase, initiate the digestive process by breaking down food when it enters the oral cavity. Fry et al. (2006) suggested that all snakes and lizards have potentially toxic components in their saliva, and the lineage may have originally been venomous. Most venoms are proteins recruited from molecules that perform different functions; simple chemical changes result in the novel protein acquiring a proteolitic or toxic activity (Fry et al., 2009). For example, the protease BLTX in Blarina converged on a nearly identical protein structure to the toxin GTX in the Mexican beaded lizard (Heloderma horridum) with very similar effects on tissue (Aminetzach et al., 2009). These observations suggest that venom enzymes evolved independently multiple times across a wide range of taxa from basic salivary enzyme building blocks.

Several studies of mammalian venom demonstrated that there are a wide variety of digestive compounds produced in saliva. Pearson (1950) found in Sorex and Cryptotis parva granular tissue similar to the venom producing cells in Blarina – their saliva had no impact on experimental animals. He also observed that estrous female B. brevicauda lacked the granular segment of the submaxillary gland, and suggested that hormones may have an impact on venom production in these animals. Hiramatsu et al. (1980) found that saliva from male mice is highly toxic to guinea-pigs, rats and hamsters, and contained kallikrein-like enzymes. They did not test female mice, but suggest that, based on gross morphology, females lack the serous-like granules that produce the compounds. Vampire bats secrete two classes of anticoagulants: plasminogen activators and inhibitors of proteinases (Basanova et al., 2002 and Ligabue-Braun et al., 2012), which are similar to the peptidase S1 toxins found in Blarina (Fry et al., 2009). Therefore, the basic morphological and chemical components to produce venom are present across a much wider range of mammalian species than produce venom, and show a large amount of variation.

We suggested previously (Folinsbee et al., 2007) that to infer venom in an extinct species, it should be closely phylogenetically related to a clade in which venom is present. However, recent studies have shown that the chemical building blocks of venom are relatively widespread, suggesting that venom acquisition may evolve more easily than previously thought (Fry et al., 2009). In mammals, this does not require a retention of the venomous condition (contra Dufton, 1992, Fox and Scott, 2005 and Hurum et al., 2006); rather, the morphological features (salivary glands, either parotid or submaxillary) and physiology (secretory proteins that digest large molecules) present in all mammals can be modified through genetic mutations to produce and deliver substances that are neurotoxic to prey.

Venom can potentially serve four functions: defence, an aid to digestion, to assist in oral hygiene, and to assist in prey capture (Arbuckle, 2009). Of these, prey capture is the most likely explanation for the presence of venom in mammals (Martin, 1981 and Tomasi, 1978).

Although the relationship between venom, BMR and body mass is not significant, Blarina and Neomys have both large body mass and high metabolic rates (Fig. 3). There may be selective pressure on a trait like venom, which enables large shrews to collect more prey in order to maintain their mass and high metabolism. Smaller shrews, even with high BMRs, do not need to consume as many calories, and may therefore be capable of storing sufficient food without the need for venom.

Alternatively, there may be a high cost to venom production, which results in selection against the evolution of venom in most animals. Although the basic capability to produce, secrete and deliver digestive enzymes of varying degrees of strength is relatively primitive and widespread across mammals, there may be significant metabolic costs to producing neurotoxic agents in salivary glands. There are no data on the metabolic costs of venom production in mammals, but data from snakes and arachnids suggest that replenishing venom incurs a relatively high metabolic cost (McCue and Mason, 2006, Morganstern and King, 2013 and Pintor et al., 2010), and that most venomous taxa limit the amount of venom they expend per bite (Morganstern and King, 2013). This alone could explain the rarity of venom across mammals.

A very small proportion of the 452 extant eulipotyphlan species have been tested for oral venom. This is partly because it is logistically difficult – large samples of fresh saliva are difficult to collect, shrews are difficult to maintain in captivity and salivary proteases degenerate rapidly. Shrews are also small, cryptic, nocturnal and tend to be endemic in small ranges. However, to confirm that venom is relatively rare in eulipotyphlans, many other species should be tested for the presence of salivary venom. In addition, understanding the genetic basis for venom in Neomys and Blarina, and the chemical differences between their venoms would provide a test of the hypothesis postulated here, that venom evolved independently in both species. If Neomys has a different toxic compound from BLTX, it would be strong support that venom evolved independently in the two taxa [although, since BLTX-like proteases evolved independently in both Blarina and the Mexican beaded lizard (Aminetzach et al., 2009), we might predict a similar toxin to be present in Neomys].

In addition, more natural history data on the rate of prey caching in venomous compared to non-venomous eulipotyphlans in the wild would greatly enhance the field. Do Blarina and Neomys cache significantly greater prey mass than other shrews that hoard prey? These venomous species preferentially focus on large prey (Rychlik, 2009 and Rychlik and Jancewicz, 2002); does this allow them to store more food overall? More data on Solenodon, including determining its basal metabolic rate, daily food intake, and foraging strategies would add enormously to this topic.

Another productive line of research would be to investigate the cost of venom production in mammals. Studies in snakes found that venom production significantly increased metabolic rate in pitvipers (McCue and Mason, 2006) and death adders (Pintor et al., 2010), and both snakes and arachnids meter their venom carefully (Morganstern and King, 2013). Assessing the metabolic cost of venom production in mammals, and determining how mammals expend venom during predation could fill gaps in our knowledge. A comparative phylogenetic approach, looking at the metabolic costs of predation in sister-pairs of venomous and non-venomous shrews, would clarify the energetic expense of venom production in mammals. If venom is metabolically costly, it would explain why the trait is not more widespread among mammals. If it is not costly, then we need to develop alternative hypotheses: perhaps mammalian teeth are used more intensively for oral processing, so there is a constraint on modifying them to inject venom; without modified venom delivery teeth, venom is less adaptive.

Ancestral character state reconstruction of the presence of salivary venom in Eulipotyphla reveals that the trait evolved in parallel in at least three distantly related lineages, and was unlikely to be present in all early mammals. This suggests that the capacity for manufacturing venom in salivary glands may be evolutionarily plastic, while the trait itself is still relatively rare. The presence of venom in described fossil shrews with grooved teeth (Cuenca-Bescós and Rofes, 2007, Furió et al., 2010 and Rofes and Cuenca-Bescós, 2009) cannot be substantiated by dental morphology alone, because no living shrew possesses both deeply grooved teeth and salivary venom. The solenodon relative Nesophontes (Turvey, 2010) also has dental morphology unlike any living venomous taxon, but its close phylogenetic relationship to solenodons and deeply grooved upper canine make the inference of venom somewhat more likely.

It is possible, as suggested by Fox and Scott (2005), that the poor preservation of small mammals from the Early Cenozoic is the reason for the lack of documentation of widespread oral venom delivery systems. However, it is more plausible that the rarity of oral venom systems found in fossil mammals is due to the extreme rarity of the phenomenon in nature. The really interesting questions are: since some mammals are capable of secreting proteins that can deter predators and immobilize prey, why is this trait so rare among mammals? If venom is energetically expensive (Morganstern and King, 2013), why is it relatively common in living snakes [∼450 of ∼2700 (17%) of snakes are venomous]?

More research into the genetic basis of venom, the chemical differences among mammalian oral venom and the natural history of venomous species would resolve these longstanding questions.