On Some Phylogenetic Aspects of Coral Snake Coloration and the. Associated Mimicry Complex

On Some Phylogenetic Aspects of Coral Snake Coloration and the Associated Mimicry Complex Kevin Arbuckle Supervisors: Prof Graeme D. Ruxton and Prof Roderic D.M. Page This report is submitted in partial fulfilment of the requirement for the degree of BSci (hons) in Zoology at the University of Glasgow in 2010

Phylogenetics of coral snake coloration - Summary Summary Coral snake mimicry has been the focus of a large research effort and much controversy over the past few decades, more than any other example of mimicry in vertebrates. Despite this, neither coral snake coloration nor mimicry has been analysed using phylogenetic methods an essential step if we are to understand the phenomenon from an evolutionary perspective. This study represents the first attempt to correct this deficit and apply molecular phylogenetics to examine coral snake coloration and mimicry. I aimed to assess how the how phylogeny has influenced the current phenotypes and how these have evolved. I also tested whether all mimic taxa evolved after the coral snakes, as would be predicted if coral snakes are the model in this mimicry complex. Phylogeny has influenced the current patterns of coral snake coloration and the latter has evolved many times independently in the mimics. The ancestral pattern of coral snakes consists of tricolour monads and the data suggests that more complex patterns are more recent innovations. More species of the family Dipsadidae are putative coral snake mimics than species of Colubridae. The former consists of (often mildly) venomous species whereas over half of Colubridae mimics are completely non-venomous. Furthermore my data suggest that some Dipsadidae mimics diverged before the coral snakes. This tentatively supports the idea of Mertensian mimicry, in which less dangerous species are models for the coral snakes (whose venom is more toxic). This idea is discussed at length, gaps in our knowledge are highlighted, and directions for future research programmes are suggested. 2

Phylogenetics of coral snake coloration - Contents Contents Acknowledgements........ 4 Introduction...5 Coral snakes and their coloration......5 Mimicry.7 Imperfect mimicry...8 Venomous snake mimicry...11 Coral snake mimicry. 12 Aims....15 Methods...16 Data acquisition.16 Taxon and gene sampling... 17 Sequence alignment....18 Phylogenetic analysis.....19 Results....... 21 Phylogenetic analyses......21 Colour pattern evolution.....23 Coral snake mimicry... 25 Discussion... 27 Phylogenetic analyses......27 Colour pattern evolution........28 Coral snake mimicry... 31 Directions for future studies....36 References... 38 Appendix 1 (Data).....46 Appendix 2 (GenBank Accession Numbers)....53 Appendix 3 (Phylogenies)....58 3

Phylogenetics of coral snake coloration - Acknowledgements Acknowledgements First and foremost I must express my most sincere gratitude to my supervisors: Graeme Ruxton and Rod Page. Their encouragement, enthusiasm, and willingness to discuss any issues that may arise was second to none, and their proof-reading skills doubtlessly helped to improve the manuscript. I also wish to thank Rod for opening the door for me to learn how to implement phylogenetic analyses using a wide variety of software programmes. Harry Greene provided me with invaluable literature and his generosity in sharing information both for this project and during previous correspondence far exceeded any expectations I harboured. Ray Hunter and Klaus Roemer kindly allowed me the use of their photographs for the title page of this thesis, and this immeasurably helped to illustrate the beauty of the coral snakes. Finally I thank Beth Williams-Arbuckle, my wife, for putting up with my almost constant attachment to my laptop for several months as I was carrying out and writing up this project. 4

Phylogenetics of coral snake coloration - Introduction Introduction Coral snake coloration (often a bright combination of red, yellow, and black) and mimicry has stimulated a large research effort and a great deal of controversy over the last few decades. This mimicry system is one of the few well-studied examples of mimicry in vertebrates, and has led to a broader understanding of how the phenomenon operates. Despite this emphasis neither coral snake coloration nor mimicry has been studied from a phylogenetic perspective. Since this would provide evolutionary insights into the mimicry complex my study represents the first attempt to correct this deficit and use phylogenetic methods to examine coral snake coloration and mimicry. Phylogenies can be used to infer the state of a given trait in common ancestors via ancestral character reconstruction (Schultz et al., 1996), to test hypotheses of co-evolution and co-speciation (Page, 2003), or to provide a basis for controlling for phylogeny in ecological studies (Felsenstein, 1985). Compared to the use of trees in the fields of systematics and taxonomy, their application to test hypotheses in evolutionary ecology is not as widely recognised. Coral snakes and their coloration The term coral snake has been used for a number of genera of venomous snakes in the family Elapidae. However the name is most closely associated with a clade of three genera of New World elapids: Micrurus, Micruroides, and Leptomicrurus. Thus the name coral snake is used here to refer to these genera. Although for consistency with previous work on mimicry I have continued to use Leptomicrurus here, previous phylogenetic studies have found this genus to be nested within Micrurus (Slowinski, 1995; Gutberlet and Harvey, 2004), and so the four Leptomicrurus species should properly be included in Micrurus. All members of the Elapidae contain potent neurotoxins in their venom and almost all are toxic enough to cause human fatalities (Roze, 1996; Mackessy, 2010). Despite this coral snake bites are responsible for only a small proportion (~1-2%) of snake 5

Phylogenetics of coral snake coloration - Introduction envenomations throughout their range, most result from viper bites (Norris, 2004; Warrell, 2004; Gutiérrez, 2010; Smith and Bush, 2010). Nevertheless a coral snake bite is a medical emergency and as a group they are of considerable importance to humans. Perhaps the most conspicuous feature of coral snakes is their coloration, and many other species have adopted similar patterns (discussed later). This has led to the need for a definition of coral snake coloration. The one adopted here is from Savage and Slowinski (1992) who define coral snake coloration as any dorsal pattern found in a species of venomous coral snake and/or any dorsal pattern containing a substantial amount of red, pink or orange distributed in such a fashion that resembles a pattern found in some species of venomous coral snake. Savage and Slowinski (1992) provided a comprehensive list of colour pattern codes devised to incorporate the entire range of coral snake coloration and presented a summary of the patterns both in coral snakes and other snake species. The distinctive patterns of coral snakes are generally considered to function as an aposematic (warning) signal. Aposematism signals to a predator that the potential prey species is dangerously well defended and thus fundamentally benefits both parties (Ruxton et al., 2004). That coral snake coloration functions as an aposematic signal (particularly for potential avian predators) has been demonstrated experimentally by several studies (Smith, 1975; Smith, 1977; Brodie, 1993; Brodie and Janzen, 1995). Criticisms of the concept of aposematism in lethal species such as coral snakes have been suggested, but all have been based on false assumptions such as strictly nocturnal habits (limiting effectiveness of visual signals) and 100% mortality of predators from bites (Greene and McDiarmid, 1981). Thus there is currently no good evidence against the use of aposematism by coral snakes and substantial evidence for it. This is not to suggest that the only possible function of conspicuousness in coral snakes is aposematism since it is 6

Phylogenetics of coral snake coloration - Introduction not mutually exclusive to others such as crypsis or disruptive coloration. However well designed experiments as cited above do strongly suggest that it is at least one function. Mimicry Mimicry is a phenomenon in which one organism dupes another into mistaking it for something else by its appearance, behaviour, or a combination of these. It most often functions as an antipredator mechanism but can also be used to attract prey (aggressive mimicry), pollinators (floral mimicry), or to increase mating opportunities (sexual mimicry) by deceiving organisms other than predators (Ruxton et al., 2004). Nevertheless I will focus on mimicry as an antipredator strategy since that is most relevant to the current study. Mimicry in this case confers the appearance of a dangerous or noxious species that would be unprofitable to a predator. Through time there have been a wide diversity of types of mimicry proposed, Pasteur (182) lists seven types of protective mimicry and 18 types in total, but most (if not all) known cases of defensive mimicry can be classified as Batesian or Müllerian. Mertensian mimicry has also been suggested to be important in snakes but this will be discussed later. In Batesian mimicry suitable prey are mimics of dangerous or noxious models whereas in Müllerian mimicry both species are dangerous or noxious and so mimic each other (Ruxton et al., 2004). The advantages to the mimic of Batesian mimicry are intuitively obvious a predator will avoid the mimic having mistaken it for unprofitable prey (the model). In Müllerian mimicry both prey species benefit since predators that learn to avoid one model will also avoid the other (note that in Müllerian mimicry both species act as the model and the mimic). This distinction seems clean cut but in reality it is best to think of a Batesian-Müllerian mimicry continuum (Speed and Turner, 1999; Balogh et al., 2008). To help illustrate how the line can be blurred, imagine an mimic which produces a distasteful and rarely nausea-inducing and a model which produces a secretion which causes more serious symptoms in a predator, occasionally even death. Clearly both species are noxious prey but one is more so than the other. Is this a case of Batesian 7

Phylogenetics of coral snake coloration - Introduction mimicry of the better defended model or are both species benefiting from Müllerian mimicry? Nevertheless it is difficult to place any given species on this conceptual Batesian- Müllerian axis and the two terms will likely continue to be used to describe species based on which extreme they fall closest to. Similarly the idea that these two types of mimicry form a continuum implies that they are fundamentally the same thing, and this is a justifiably practical way of thinking. However, Ruxton et al. (2004) suggested a more technical distinction based on the selective forces used to generate the mimetic association. Under these definitions Batesian mimicry has evolved as a result of selection to deceive predators in thinking the mimic is unprofitable in some way, in contrast Müllerian mimicry has evolved from selection to spread the costs of predator learning of the aposematic signal. There are theoretical benefits to engaging evolutionary concepts in such definitions but in reality it is difficult to separate the hypotheses generated by these different definitions and difficult to reach a classification based on selection pressures in cases that seem functionally intermediate. As such it is likely that the traditional functional distinctions will continue to be more commonly used in studies of mimicry assemblages. Imperfect mimicry Rarely are mimics perfect since details of the colour patterns (for example) are different from the model. If it is similarity to the model that confers the antipredator advantage then we might expect selection to continue to improve the likeness until the mimic is a perfect representation of the model. At least five hypotheses can be invoked to explain the phenomenon of imperfect mimicry, though these are not mutually exclusive. 1) Perfect mimicry is simply not needed since the discrimination abilities of predators are not perfect. Sherratt (2002) found that mimicry is effective over a relatively broad range of phenotypes around that of the model so that small deviations from the 8

Phylogenetics of coral snake coloration - Introduction exact phenotype of the model are not under strong selection pressure to improve the resemblance. 2) Sherratt (2002) also found that imperfect mimicry can evolve where the mimic overlaps more than one model species. In this case selection may actually favour an intermediate phenotype in order to gain protection from the mimicry of more than one source (assuming both models are broadly similar in appearance). 4) Batesian mimicry is a relatively unstable strategy since it is in the interest of the model to evolve away from the signal being exploited by the mimic, but not so far that it becomes unrecognisable itself and loses the benefits of aposematism (Endler, 1991). Therefore an arms race is set up whereby it is difficult for the mimic to ever match the model exactly. This is particularly the case if the mimics exist at a high frequency relative to the model. In this scenario predators can learn that prey with an aposematic signal can be eaten some of the time (when the prey is the mimic, not the defended model) and thus the fitness of the model is decreased. Models that are particularly dangerous or noxious may be able to resist such effects from high mimic densities since the cost to a predator of mistakes is high and so outweighs the benefit of obtaining a meal (Johnstone, 2002). 5) Generalisation by predators can reduce the requirement for perfect mimicry in a similar way to the poor discrimination hypothesis. If predators only use certain cues from the aposematic signal then providing these are present in the mimic the exact nature of the entire signal is not important. This appears to be the case for all studies that have allowed individual cues to be assessed and presumably buffers the predator from natural variations in the model. Chicks that were trained to associate a distasteful substance with beads painted chrome, black, yellow, or black and yellow only showed subsequent avoidance to the combination of black and yellow (Johnston and Burne, 2008), suggesting that the important cue in this system is the combination of yellow and black, not individual colours. 9

Phylogenetics of coral snake coloration - Introduction Kauppinen and Mappes (2003) found similar results using dragonflies as predators, which predate flies but avoid wasps. When flies were painted black and yellow they were attacked less than if painted either black or yellow. Odours and body shape had no effect, indicating again that it is the combination of black and yellow that acts as the cue in this system. Therefore any mimics that achieved black and yellow coloration may not need perfect resemblance to the model (wasps in this case) in order to benefit. Finally, a different cue appears to be important in ladybirds. Dolenská et al. (2009) found that colour itself did not affect predation by a bird predator but that the presence of spots reduced attacks significantly. Clearly if the presence of spots is all that is required for this aposematic signal to be recognised by predators then imperfect mimicry would not pose a problem providing the mimics could evolve a spotted pattern. Similarly this might also favour a diverse assemblage of Müllerian mimics in ladybirds without necessitating a high degree of convergence in pattern. Mimicry is a taxonomically widespread antipredator mechanism. There are suspected or confirmed examples of Batesian mimicry in insects and other invertebrates, amphibians, snakes, fish, and even between reptiles and invertebrates, birds or mammals (Rettenmeyer, 1970; Howard and Brodie, 1973; Sánchez-Herrera et al., 1981; Pough, 1988; Roze, 1996; Greene, 1997; Brodie and Brodie, 2004; Ruxton et al., 2004; Wüster et al., 2004; Green and McDiarmid, 2005; Brown, 2006). Similarly, Müllerian mimicry has been reported for many groups of insect, snakes, and frogs (Rettenmeyer, 1970; Roze, 1996; Brodie and Brodie, 2004; Ruxton et al., 2004; Greene and McDiarmid, 2005; Sanders et al., 2006; Baxter et al., 2010). Perhaps in part because of this diversity mimicry has received a large research effort (though mostly on Heliconius butterflies; Brodie and Brodie, 2004), but very few studies have used a phylogenetic approach (Ruxton et al., 2004; but see Sanders et al., 2006; Bocak and Yagi, 2010) and none have used this to address venomous snake mimicry (Greene and McDiarmid, 2005). 10

Phylogenetics of coral snake coloration - Introduction Venomous snake mimicry Some workers have suggested that venomous snake mimicry may have unique attributes not found in insect mimicry (Pough, 1988; Brodie and Brodie, 2004). If this is true then it opens up a large amount of research opportunities since most studies of mimicry have focussed on insects (particularly butterflies). Mimicry of venomous snakes by other snakes is extremely common with 25-35% of harmless species using this strategy (Greene and McDiarmid, 2005). Many of the features proposed to be unique to venomous snake mimicry have been discussed by Pough (1988) who list three basic characteristics. 1) Venomous snakes represent highly dangerous models and so should lead predators to generalise the aposematic signal of the model. This in turn can result in a high degree of abstract mimicry in which no specific model can be identified and only a generalised resemblance is achieved. Abstract mimicry is therefore similar to imperfect mimicry and appears to occur to a higher degree amongst snakes than insects. Recent modelling studies have supported the suggestion that more dangerous models confer better protection to imperfect mimics (Johnstone, 2002; Sherratt, 2002). Abstract mimicry resulting from this generalisation may also partly explain the phenomenon of other groups of organisms such as insects (especially caterpillars) mimicking venomous snakes (Pough, 1988; Brown, 2006). 2) Different predators may use different cues from the aposematic signal so that by combining cues from different model species (dual mimicry) a mimic may obtain protection from a wider variety of predators. 3) The highly dangerous nature of venomous snakes may also exert strong selection pressure on predators to avoid the models since an encounter can lead to mortality or morbidity (loss of limbs etc.) both of which have negative fitness consequences. The larger size of snake models compared to insect models both absolutely and relative to predators probably also makes them more formidable prey. This likely simplified the evolution of mimicry in other taxa and may explain the high diversity of venomous snake mimics. 11

Phylogenetics of coral snake coloration - Introduction Mertensian mimicry also appears to be unique to venomous snakes and involves a lethal snake mimicking a less dangerous model. This has been a controversial hypothesis since it relies on an absence of learning opportunities for predators of the highly venomous snake (Brodie and Brodie, 2004; Greene and McDiarmid, 2005). Mertensian mimicry will be discussed in more detail later in this thesis. Greene and McDiarmid (2005) proposed four macro-evolutionary consequences of snake mimicry which they termed Savage-Wallace effects. Firstly, mimicry is more likely to occur among closely related organisms sharing a common body plan (e.g. among snakes). Species that are only distantly related and as a result differ dramatically in general morphology are less likely to form a mimicry complex since predators are less likely to be deceived. Secondly, where mimicry does involve distantly related organisms it will tend to evolve among those with a similar (usually relatively simple) body plan, for example between snakes, caterpillars, myriapods, or flatworms. Third, as a consequence of the first and second effects venomous snake mimicry is unusually widespread, this is also influenced by the high potential costs to a predator of attacking highly dangerous prey. Lastly, the origin of noxious or dangerous defences (e.g. venom) can increase the diversity of clades other than those possessing the defence. This occurs since mimicry of these dangerous models makes unprotected niches available for harmless species that would be unable to exploit them otherwise. Though interesting and generating many hypotheses, the Savage-Wallace effects appear to have received no study to date and represent fertile ground for future research programmes in venomous snake mimicry. Coral snake mimicry Almost 20% of New World snakes are considered coral snake mimics (Savage and Slowinski, 1992), and this highlights the extent of the mimicry complex surrounding this group. Furthermore, various non-snake taxa are suspected of being coral snake mimics including a caterpillar and a turtle (Roze, 1996), though neither of these are confirmed as such. Conversely, at one time it was thought that the coral 12

Phylogenetics of coral snake coloration - Introduction snakes may actually be mimicking noxious and aposematic millipedes, however experimental studies do not support this hypothesis (Brodie and Moore, 1995). Coral snake mimicry has received more attention that any other mimicry complex involving vertebrates and has also generated considerable controversy. Using naïve avian predators (motmots and kiskadees, Eumomota superciliosa and Pitangus sulphuratus respectively) and painted wooden models, Smith (1975, 1977) demonstrated that the avoidance of coral snake coloration is an innate characteristic and so can provide protection without having to be learned. The birds readily pecked at models that were a single colour (including red or yellow) or ringed with green and blue, but did not peck at yellow and red ringed models. Red and yellow striped models were also attacked in this study, but not as frequently as controls. Similarly plasticine models have been used to test responses of free-ranging predators to coral snake coloration. Brodie and Janzen (1995) found that birds frequently attacked plain brown models but avoided those with both bicolour and tricolour coral snake patterns. Brodie (1993) also conducted a similar experiment using a greater variety of colour patterns and with models on a natural or white background to remove any effects of crypsis. Regardless of background the brown (control) model was attacked more often than a tricolour coral snake model. Furthermore the brown model was attacked significantly more than all models exhibiting coral snake coloration but there was no difference in other (non-predator) damage sustained by the models, indicating that it is indeed selective predation that caused more attacks on the control model. Notice that all these studies have used birds as predators, likely because birds tend to be visually orientated predators compared to mammals or reptiles which use chemical senses. Beckers et al. (1996) using a mammalian predator (a coati, Nasua narica) that was temporarily kept in captivity and presented with a variety of live snakes including coral snakes, mimics, and both venomous and harmless cryptic taxa. Interestingly the coati 13

Phylogenetics of coral snake coloration - Introduction showed no avoidance to species with coral snake coloration. It could be that wooden or plasticine models are an inappropriate method of testing but given that differences are observed in studies using them and it eliminates all cues except colour this is unlikely. It is probably related to birds being more visual predators compared to mammals and therefore mimicry and aposematism may be more effective against birds. A given defence does not have to be effective against all possible predators to provide some advantage and hence to be selected for. Some of the most convincing evidence for coral snake mimicry relates to parallel changes in colour pattern of mimics and coral snakes across their range and with ontogeny. Green and McDiarmid (1981) provide important examples of Pliocercus and Erythrolamprus (both mimics) exhibiting different patterns throughout their range and those patterns being very similar to the local coral snake species. The concordance of Pliocercus patterns with coral snakes is so accurate that it has been said that by finding one of these species it is possible to know the colour pattern of the local coral snake species (Roze, 1996). Savage and Slowinski (1996) reported similar findings for another mimic genus, Scaphiodontophis, and Roze (1996) has reported yet more examples including Atractus elaps. There are a number of mimic species that are only mimics as juveniles, such as Lampropeltis triangulum (some subspecies), Oxyrhopus petola, and all species of Clelia (Savage and Slowinski, 1992; Roze, 1996; Greene and McDiarmid, 2005). These ontogenetic changes appear to be related to body size since there are few large coral snake mimics (Roze, 1996; Pyron and Burbrink, 2009b). This makes sense in the context of deceiving predators since most coral snakes are also relatively small and so mimicry may only be effective until a given species is noticeably larger than the coral snake(s) acting as its model. 14

Phylogenetics of coral snake coloration - Introduction Aims Since coral snake mimicry has never been examined from a phylogenetic perspective, despite calls for more studies of mimicry using this approach (Ruxton et al., 2004; Greene and McDiarmid, 2005), this study represents the first attempt to apply phylogenetic techniques to this mimicry complex and to the evolution of coral snake coloration in both the coral snakes and their mimics. I aimed to assess how these patterns have evolved and how phylogeny has influenced the current phenotypes. I also used the phylogeny of the group to test the traditional scenario of coral snake mimicry, i.e. that coral snakes are the models and all other taxa are the mimics. This situation predicts that all the putative mimics have evolved after the coral snakes since in any mimicry complex the model must be present before the mimic. This prediction was tested here to assess whether the traditional view of coral snake mimicry is accurate. I also assessed whether the degree of polymorphism in mimics was correlated with the number of coral snake species they overlap in distribution, as might be expected if mimics have adapted their colour patterns to the local coral snake species throughout their range. 15

Phylogenetics of coral snake coloration - Methods Methods Data acquisition Colour patterns were coded using the system of Savage and Slowinski (1992), summarised in Fig. 1. Although the codes themselves give no real indication of what the pattern actually looks like there are some generalities such as codes beginning with B or T are bicolour or tricolour patterns respectively. Since I have used this coding system in subsequent discussions Fig. 1 provides a visual representation of the patterns. Savage and Slowinski (1992) also served as the primary source for determining which Figure 1 - Coding scheme used for colour patterns, adapted from Savage and Slowinski (1992). The patterns shown above are portrayed as a lateral view of part of the snake s body, with the top of the diagrams corresponding to the dorsolateral area and the bottom corresponding to the ventrolateral area. 16

Phylogenetics of coral snake coloration - Methods species can be regarded as mimics (see introduction for details of how mimicry was defined), although some minor inclusions were made which appear to have been omitted by those authors. These inclusions were based on the species exhibiting coral snake coloration (as defined in the introduction). Coral snake colour patterns were supplemented with data from more recent summaries of the group (Roze, 1996; Campbell and Lamar, 2004). For a given species I have separated different patterns present on different individuals with a comma and different patterns present on a single individual are connected with an ampersand (&). Distribution data were obtained at the level of resident countries. Mimic distributions were acquired primarily by searching the JCVI/TIGR Reptile Database (Uetz and Etzold, 1996; Uetz et al., 2007), a comprehensive online source for the taxonomy and distributions of reptiles. Coral snake distributions were primarily taken from Roze (1996) and Campbell and Lamar (2004), and these were then supplemented using the JCVI/TIGR Reptile Database. The full dataset of pattern codes and distributions is included in Appendix 1. Gene sequences were downloaded from GenBank (the standard and searchable database where gene sequences are deposited and stored) via the Query Databanks tool in MEGA version 4.1 (Tamura et al., 2007; Kumar et al., 2008). Searches were carried out using the generic names of the taxa being examined and often the name of the gene sequence desired. The GenBank accession number of each downloaded sequence was recorded for ease of future location, and these are given in Appendix 2. Taxon and gene sampling Since coral snake sequences were the prime requirement for this study I first searched GenBank to determine which gene sequences were available for coral snakes. This was the basis for the selection of genes to use for analysis, though availability of sequences for mimics also had some influence on the final selection. The genes chosen as a result of this process were c-mos, cytb, and ND4. The latter two are mitochondrial genes commonly used for systematic studies of reptiles, while 17

Phylogenetics of coral snake coloration - Methods c-mos is a nuclear gene that has also been shown to be useful in reptile phylogenetics (Saint et al., 1998). All of these genes had sequences available for a number of coral snakes including at least one representative of Micrurus, Leptomicrurus, and Micruroides. Since Micruroides is the most basal of the American coral snakes, this representation encompassed a phylogenetically broad sample of this group. For each gene, sequences from all available coral snakes were downloaded. Thereafter GenBank was searched for each mimic genus in turn and sequences were downloaded where available for mimic species and also congenerics. Furthermore, at least one outgroup (taxon outside but closely related to the group of interest) sequence was downloaded for each mimic group. Outgroup selection was based on previously published phylogenies (Pinou et al., 2004; Gower et al., 2005; Slowinski and Lawson, 2005; White et al., 2005; Kelly et al., 2009; Pyron and Burbrink, 2009a; Zaher et al., 2009), and these taxa were used to root all phylogenies constructed in this study. Outgroup rooting is a standard method of determining the evolutionary sequence of the phylogeny. Sequences were available for the following groups (plus the relevant outgroups): coral snakes, Anilius, Dipsadidae, Farancia, Lampropeltini, and Leptodeira. Furthermore, composite trees were constructed for each gene using all available coral snake, mimic, and outgroup sequences. The term composite trees is used here to refer to phylogenies combining all taxa available for a given gene, in contrast to the trees for individual sequence groups (e.g. coral snakes or Leptodeira). Sequence alignment All alignments were carried out using ClustalW (Thompson et al., 1994) as implemented in MEGA version 4.1 (Tamura et al., 2007; Kumar et al., 2008). Alignments were carried out on the translated protein sequences since this method is more accurate than aligning the sequences directly (Hall, 2005). This is an important consideration since it is the alignment that is the data used for tree construction, not necessarily the sequences themselves therefore the resulting tree can only be as 18

Phylogenetics of coral snake coloration - Methods reliable as the alignment on which it is based. Pairwise alignments were performed using a gap opening penalty of 10 and a gap extension penalty of 0.1, and multiple alignments had a gap opening penalty of 3 and a gap extension penalty of 1.8, as recommended by Hall (2008). Subsequent to ClustalW alignments I made manual adjustments by eye to avoid obviously wrong alignments resulting from the mathematical (not biological) nature of the Clustal algorithm. Average amino acid identities were computed for all alignments to ensure the reliability of the latter was suitably high. Phylogenetic analysis Phylogenies were constructed by maximum parsimony (MP) and Bayesian inference (BI) for all groups of sequences for each gene. These methods were chosen to represent a likelihood approach (BI) and a non-likelihood based method (MP). All stages of the analysis represent standard and suitable methods in phylogenetics. MP trees were constructed in MEGA version 4.1 (Tamura et al., 2007; Kumar et al., 2008) using the close-neighbour-interchange algorithm with search level 7. The initial trees were obtained by random addition of sequences with 10 replicates. The complete deletion option was used so all positions containing gaps and missing data were eliminated. The reliability of the Dipsadidae MP tree was tested with 2000 bootstrap replicates, all other groups were tested with 10,000 replicates. Dipsadidae MP analysis was originally set to 10,000 bootstrap replicates but the excessive time required meant that reducing the number of replicates to 2000 was beneficial with little loss of effectiveness. BI trees were constructed using Markov Chain Monte Carlo randomisation in MrBayes version 3.1.2 (Huelsenbeck and Ronquist, 2001; Ronquist and Huelsenbeck, 2003) for all sequence groups and also for the composite trees. The programme was set to estimate different substitution rates for each codon position and to use the general time reversible model of nucleotide substitution. All parameters were estimated independently by MrBayes. All analyses were run for 1,000,000 generations with trees sampled every 19

Phylogenetics of coral snake coloration - Methods 200 generations. Four independent chains (three heated and one cold) were run with a temperature of 0.2 to allow the switching process, thereby avoiding being trapped on a local maximum far below the true maximum. For each analysis a 50% consensus tree was computed, with the first 1250 trees discarded as burnin to prevent inaccurate preconvergence trees contributing to the consensus trees. Composite trees were deemed to best allow testing of the hypotheses since all possible taxa were included on one tree and so topology and branch length for all taxa were calculated in the same analysis. These were examined by eye and since no obvious deviations from a molecular clock were found they were linearised (using MEGA version 4.1) to permit better comparisons across taxa and visualisation of relative divergence times. 20

Phylogenetics of coral snake coloration - Results Results Phylogenetic analyses Since the linearised trees best allow comparisons of relative divergence times they are presented in Figs. 2-4 and will be the main source of subsequent discussion. All other trees constructed in this study are also included in Appendix 3 for reference, and some parameters of the MP analyses for each sequence group are provided in Table 1. The topologies of all three linearised trees were similar to previously published phylogenetic hypotheses, and the major expected groupings in each tree were recovered. The overlap in taxa between the cytb and ND4 trees permits comparison between them. There is a high degree of congruence between these two trees. This suggests that the trees obtained in this study are reasonable estimates of the phylogenetic history of the group and can be used confidently as the basis for interpreting the other results obtained herein. Polytomies were obtained for a number of clades, particularly the internal relationships of the Dipsadidae. This group comprises an extensive radiation with numerous mimics that was examined using the c-mos gene. Despite this, resolved clades within the Dipsadidae often had high support in this study (Fig. 2, also see Appendix 3). Total number of sites Parsimony informative sites Length of most parsimonious tree(s) Number of equally Sequence Group parsimonious trees Anilius c-mos 546 4 2 57 coral snake c-mos 534 1 3 13 Dipsadidae c-mos 450 49 2215 137 coral snake cytb 642 95 1 317 Lampropeltini cytb 1115 260 2 979 Leptodeira cytb 713 117 1 303 coral snake ND4 657 201 5 803 Farancia ND4 844 58 1 186 Lampropeltini ND4 844 196 2 735 Leptodeira ND4 681 136 1 396 Table 1 Some parameters of MP analyses constructed in this study for individual sequence groups. The trees are presented in Appendix 3. 21

Phylogenetics of coral snake coloration - Results Figure 2 - Annotated and linearised BI tree of c-mos sequences used in this study. Node values are Bayesian posterier probabilities and branches with support <50 were collapsed. Filled circle represents coral snake clade while underlined taxa are mimics. Text following focal taxa is arranged as: [pattern code(s)] [country(ies) included in distribution]. 22

Phylogenetics of coral snake coloration - Results Figure 3 - Annotated and linearised BI tree of cytb sequences used in this study. Node values are Bayesian posterier probabilities and branches with support <50 were collapsed. Filled circle represents coral snake clade while underlined taxa are mimics. Text following focal taxa is arranged as: [pattern code(s)] [country(ies) included in distribution]. Colour pattern evolution The coding system employed here is not amenable to formal mapping of ancestral states (to determine the ancestral colour patterns in each case), however it seems that coral snake coloration (see introduction for definition) has evolved independently at least 12 times (excluding in the coral snakes themselves). Certainly this trait has evolved on multiple occasions. The most comprehensive tree of coral snake taxa was obtained for ND4 (Fig. 4) and shows a large clade of tricolour triad 23

Phylogenetics of coral snake coloration - Results Figure 4 - Annotated and linearised BI tree of ND4 sequences used in this study. Node values are Bayesian posterier probabilities and branches with support <50 were collapsed. Filled circle represents coral snake clade (extension to include Micruroides) while underlined taxa are mimics. Text following focal taxa is arranged as: [pattern code(s)] [country(ies) included in distribution]. 24

Phylogenetics of coral snake coloration - Results (TT) species. The two most basal coral snake clades (Micruroides euryxanthus and Micrurus corallinus + Micrurus fulvius) both show a tricolour monad (TM) pattern, indicating that this may be the ancestral state. There are few instances where mimic taxa form clear clades, however the Lampropeltis group (including Cemophora and possibly Rhinocheilus) appears to have evolved mimicry early in its radiation (Figs. 3 and 4). Within the Dipsadidae Clelia + Drepanoides appears to be the only clear example of a multi-generic mimic clade (Fig. 2), but the poor resolution of the internal relationships may render this an underestimate. This latter clade consists of unicolour snakes with a black nuchal collar (Ucd). Amongst the Lampropeltis group colour patterns are highly variable but there does appear to be a clade of tricolour diad (TD) species containing the following taxa (relationships between which are variably supported): L. zonata, L. webbi, L. pyromelana, L. t. arcifera, L. t. conanti, L. ruthveni, L. mexicana. Although I have designated this a TD clade there are two exceptions contained within: L. webbi has reverted to a non-mimic state and L. mexicana is polymorphic but does not include TD in its array of forms. These are presumably autapomorphies in these species. While the most complex array of colour patterns (L. mexicana) is a comparatively recently derived species, other complex patterns are found in species at more basal species in the context of the Lampropeltis group. Coral snake mimicry Mimic taxa (or clades) in both the cytb and ND4 trees are more recent than the sympatric coral snake species (Figs. 3 and 4). In contrast, the Dipsadidae have some mimic taxa that have seemingly diverged before the main coral snake radiation regardless of degree of sympatry including Lystrophis histricus, Lystrophis dorbignyi and Erythrolamprus aesculapii (Fig. 2). These predivergence taxa in some cases display typical coral snake patterns (e.g. Lystrophis histricus displaying tricolour monads). Predivergence taxa is used here to refer to mimic taxa that seem to have evolved before coral snake taxa based on the phylogenies herein. 25

Phylogenetics of coral snake coloration - Results Of the 116 accepted mimic taxa (those covered in this study) ca. 39% are polymorphic as compared to the 76 coral snakes of which ca. 25% are polymorphic. Furthermore, 89 of 709 (13%) members of Dipsadidae are mimics, far higher than Colubridae (26 of 653; 4%)(Yates corrected χ 2 1=8.62826, P=0.0033). All coral snake mimics fall into these two families with the exception of Anilius scytale which is in the monotypic Aniliidae. The absolute number of sympatric coral snake species did not explain the number of polymorphic states (number of colour patterns displayed by a species). However when both variables were controlled for phylogeny using the independent contrasts/species pair method, a generalised linear model (based on a Poisson distribution) showed a significant (though weak) association between the number of sympatric coral snake species and polymorphic states (GLM 1, 15, P=0.0006; Fig. 5). Figure 5 - Number of polymorphic states in the mimic plotted against the number of sympatric coral snake species. Both variables are corrected for phylogeny. See text for further details. 26

Phylogenetics of coral snake coloration - Discussion Discussion At least three factors can be identified that would have improved this study and will be useful for making further developments in this area. Firstly, gene sequences are only available for a relatively small proportion of coral snakes (21%) and mimic species (22%). Furthermore, sequences common to both are also uncommon and therefore there are comparatively few available data that allow the construction of phylogenies for testing hypotheses related to mimicry. Ideally comparable sequences would be available for all relevant taxa but realistically this is unlikely to happen in the foreseeable future due to the difficulty of finding all species for sampling. If at least the majority of coral snakes and mimics had suitable sequences available the resulting phylogeny would allow more rigorous testing. The development of a new coding system for the colour patterns would enable a wider range of interesting questions to be addressed. The system proposed by Savage and Slowinski (1992) is comprehensive and excellent for a variety of purposes such as cataloguing patterns and making direct comparisons. However a new system that is amenable to ancestral character reconstruction would enable ancestral appearances to be inferred and more detailed hypotheses of pattern evolution to be evaluated. Finally, the application of a molecular clock estimate to the phylogenies would provide better inference of divergence times. The approach taken here of using linearised trees was deemed acceptable since no obvious deviations from a molecular clock were observed and I was concerned with relative, not absolute, divergence times. Despite this, dating the divergence of taxa would allow firmer conclusions than the method used here. Phylogenetic analyses In general there was a high degree of congruence between trees constructed in this study (where possible) and also good concordance with previous phylogenetic hypotheses (Gutberlet and Harvey, 2004; Pinou et al., 2004; Gower et al., 2005; Slowinski and Lawson, 2005; White et al., 2005; Kelly et al., 2009; Pyron and 27

Phylogenetics of coral snake coloration - Discussion Burbrink, 2009a; Zaher et al., 2009). Combined, these observations suggest that the phylogenies obtained in this study are reliable enough to test hypotheses of colour pattern evolution and mimicry. The resolution within the Dipsadidae was relatively poor in this study and likely constrained the interpretations that could be made of evolutionary patterns within the group. Since the Dipsadidae include a large number of mimics a well resolved phylogeny would undoubtedly provide further insights into the details of the mimicry complex with coral snakes. Although I only used one nuclear gene (c-mos) in my analyses of Dipsadidae a previous phylogeny based on one nuclear and two mitochondrial genes also failed to achieve high resolution (Zaher et al., 2009), despite being the most comprehensive analysis of this group to date. The systematics of neither the coral snakes nor the mimics were the focus of this study, however the consistently poor resolution of the Dipsadidae deserves comment. This exclusively New World family contains over 700 species of mildly venomous snakes, at least 31 of which have caused clinical envenomation in humans (though this is likely an underestimate of the number capable of causing symptoms of envenomation) (Warrell, 2004). Since some of these are or have the potential to be medically important and also are important to understanding coral snake mimicry, further efforts should be made to resolve the phylogenetic relationships of this group. Increased taxon sampling and the use of either a larger number of gene sequences or perhaps different genes with a more appropriate rate of evolution may help in this pursuit. Colour pattern evolution Although formal ancestral state reconstruction of colour patterns was not possible due to the coding system adopted here some inferences of pattern evolution can be made based on the phylogenetic distribution of traits. Coral snake coloration is a feature of all American coral snake species, as the name would suggest, and tricolour monads (TM) appear to be the ancestral pattern. Other patterns are present 28

Phylogenetics of coral snake coloration - Discussion in coral snakes but there is a predominant clade of tricolour triad (TT) species, which is consistent with previous results (Gutberlet and Harvey, 2004). This indicates that there may be a strong phylogenetic component to colour patterns in coral snakes, though broader taxon sampling (especially of taxa with different patterns) would help to confirm this. The more complex tricolour triads presumably evolved from the ancestral tricolour monad pattern via the development of back rings between the red and yellow rings. Unfortunately until more sequences are available for the construction of a more inclusive phylogeny it will remain unknown how other coral snake patterns fit into this paradigm or how they otherwise evolved. Similarly, the generality of the observation that more complex patterns are more recent innovations in coral snakes will also require more data to test adequately. Although published examples that are open to testing the increasing complexity of colour patterns are rare, it does appear to hold true to some extent for a genus of Southeast Asian beetles (Bocak and Yagi, 2010), Pachytriton newts (Wu et al., 2010), Ithomia butterflies (Jiggins et al., 2006), and East African cichlids (Seehausen et al., 1999). Therefore it would not be surprising if complex patterns were also more recently evolved in coral snakes. Coral snake coloration has evolved repeatedly in many different lineages other than coral snakes. Based on the available data it has likely evolved at least 12 times independently. However with the paucity of available mimic sequences and the poor resolution of the Dipsadidae (which contains many mimic species) this is likely a substantial underestimate and it is impossible to give a more definitive number. Since this study was concerned with the phylogenetic perspective all subsequent discussion will be restricted to those species included in the phylogenies presented herein. This is an unfortunate constraint since potentially informative species did not have gene sequences available for analysis and serves to highlight the need for collecting more genetic data. As 29

Phylogenetics of coral snake coloration - Discussion an example it is interesting to note that the Costa Rican endemic Leptodeira rubricata exhibits a bicolour pattern. Despite the majority of coral snakes having a tricolour pattern three of the four species in Costa Rica are bicolour, but none of these had data available for phylogenetic analysis to shed further light on observations such as this. Clades consisting of a number of mimic species were rare in my analyses but again I suspect more will be found with better taxon sampling and resolution. Two apparent mimic clades were recovered: Lampropeltis+Cemophora and Clelia+Drepanoides. The latter consists of the monotypic Drepanoides anomalus and all 11 species of Clelia, which are only mimetic as juveniles. This may be a consequence of body size since Pyron and Burbrink (2009b) found that mimicry of venomous snakes in the Lampropeltini clade was related to small body size. Therefore in Clelia+Drepanoides all contained species are putative mimics, in contrast to Lampropeltis+Cemophora. This variation may prove useful in testing mimicry-related hypotheses but it is limited in that only two contained species are non-mimics. Based on the phylogenies presented here Rhinocheilus lecontei (another mimic) may also be a member of the Lampropeltis+Cemophora clade but its exact position was unresolved. Rodríguez-Robles (1999) did not find such an association but other studies have found that Rhinocheilus lecontei+arizona elegans (a non-mimic species) form the sister group to Lampropeltis+Cemophora (Pyron and Burbrink, 2009a, 2009b), in which case the mimic clade could be expanded to include Rhinocheilus. For the purposes of this discussion I have adopted the more conservative approach of excluding Rhinocheilus and considering the mimic clade as being Lampropeltis+Cemophora. Within this clade there is a wide variety of colour patterns (Figs. 3 and 4), but a clade of tricolour diad (TD) species is also observed. This TD clade contains the following Lampropeltis species: mexicana, pyromelana, ruthveni, triangulum, and zonata. It should be noted that of these species zonata is polymorphic but its range of patterns includes TD, 30