Study Organism

The Western Rattlesnake (Crotalus viridis) is a polytypic North American pit viper [16, 23–24], widely distributed across broad latitudinal and elevational gradients. It extends from the Missouri River in the east, into Saskatchewan, Alberta, and British Columbia to the north, along the west coast of the United States, and south into Mexico (Fig 2; [23]). Habitat includes deciduous and coniferous forests, scrub, prairie grasslands, shrub steppe, desert margins, and sand dunes as arrayed across a gradient from sea level to 4000m [24]. As a consequence, it displays considerable variation within and among populations, life stages, and subspecies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The topographic distribution of nine Western Rattlesnake OTUs (Operational Taxonomic Units) in North America.

The nine Western Rattlesnake OTUs vary greatly in size and extent of distributional area. The Colorado Plateau represents a suture between eastern and western lineages, as six of nine occur there as either parapatric or sympatric.

https://doi.org/10.1371/journal.pone.0146166.g002

The traditionally accepted taxonomy is a single, polytypic complex (Crotalus viridis) defined largely via geography, meristics, and an assumption of intergradation. It is composed of nine OTUs, and the sample sizes for our GM analyses are indicated in parentheses following each taxonomic designation: C. v. abyssus (N = 31); caliginis (N = 33); cerberus (N = 163); concolor (N = 144); helleri (N = 446); lutosus (N = 515); nuntius (N = 256); oreganus (N = 590); viridis (N = 646). Molecular data suggests the presence of three species: C. viridis (to include viridis and nuntius); C. oreganus (to include abyssus, caliginis, concolor, helleri, lutosus, oreganus); and C. cerberus [17].

Acquisition of Shape Data

High-resolution digital photographs of the dorsal and lateral head were obtained from 3,170 adult, juvenile, and neonate museum specimens of Western Rattlesnake (S1 Table) using a Nikon D90 digital single lens reflex camera and a Nikon 105mm f/2.8G ED-IF AF-S VR Micro-Nikkor Lens. Of these, 2,824 (89.1%) were quantified, effectively spanning the range of each as an OTU (operational taxonomic unit).

A total of 33 dorsal and 50 lateral homologous landmarks were identified and digitized on 2,824 individuals using the tpsDIG2 program [25] (Fig 3). Landmarks consisted of both “fixed” and “sliding” semi-landmarks [26–27], defined by their Cartesian coordinates. Generalized Procrustes Analyses (GPA; [28–29]) were applied to equilibrate landmark configurations with regard to size, orientation, and position, such that variation among landmarks was attributed only to variation in shape. This is a critical yet often overlooked consideration in GM analyses, in that it promotes the use of collections and specimens that vary in age and method of preservation. Minimum bending energy was used to slide landmarks, and their configurations were then aligned and centered by GPA (and termed “Procrustes residuals”) so as to depict individuals in “shape space.” Values from this non-Euclidean space were then projected into its tangent space so as to yield shape variables applicable for statistical analyses [30–31]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Lateral and dorsal landmarks used for subsequent shape analysis in the Western Rattlesnake complex.

A total of 33 dorsal and 50 lateral landmarks were selected to assess shape in the Western Rattlesnake complex. Landmarks were composed of both Type I (white) and sliding semi (black) landmarks.

https://doi.org/10.1371/journal.pone.0146166.g003

The lateral landmark configurations included anatomical landmarks associated with an articulated structure, the mandible. This angle of articulation must be accommodated so as to eliminate its respective non-shape landmark variation in the overall configuration. To do so, we applied the “fixed angle” method [32] that aligns the angle of articulation (arbitrarily set at 40°) prior to GPA.

Additionally, after GPA, we sought to determine if lateral and dorsal landmark configurations could be considered integrated subsets of a combined head shape. To test for morphological integration, we performed a two-block partial least squares (PLS) analysis between dorsal and lateral landmark configurations [33]. PLS indicated significant correlation (r_PLS = 0.5485, P = 0.0001), suggesting dorsal and lateral shapes were integrated and an independent analysis of each would be inappropriate. Therefore, dorsal and lateral landmarks were combined into a single data set by applying the modified “separate subsets” method [32]. Each was scaled by the ratio of the subset centroid size to the sum of both for each subject; i.e., where subscripts _D, _L, and _T, correspond to dorsal, lateral, or total components, respectively, of either the centroid size (CS) or the vectorized matrix of Procrustes residuals (y) (i.e., each y^t is a row vector of specimen Procrustes residuals that can be compiled into a data matrix.) Thus, combined Procrustes residuals represent a configuration with unit centroid size.

Procrustes residuals were projected onto the first two PCs to visualize shape variation. A thin-plate spline (TPS; [34]) function was used to create “transformation” grids [35] that depicted the mean shape associated with different locations in the shape space (as projected into the PCs of the tangent space). We performed TPS on lateral and dorsal configurations separately.

Data Analyses

Potential sources of head shape variation included species or subspecies designation (i.e., mutually exclusive models of genetic relatedness), sex, stage of development (juvenile or adult), and head size. We constructed all possible linear models that included each of the factors (species or subspecies, sex, stage) and head size as a covariate, with factor-covariate interactions. Head size was calculated as the log of the centroid size—the square root of the summed squared distances between anatomical landmarks and the centroid of the configuration—for the dorsal landmark configuration [36]. We performed a method of model selection (S1 Text) that identified head size and subspecies as sources of variation, but not their interaction, suggesting a common shape allometry among subspecies [37–42].

We performed a non-parametric (np) multivariate analysis of variance (MANOVA) on the common-allometry model using a randomized residual permutation procedure (RRPP) so as to test model effects and gauge their sizes [43]. This procedure also permitted the calculation of pairwise distances between least squares means for subspecies in every random permutation, and the uniqueness of observed subspecies shapes to be tested. We performed np-MANOVA with these pairwise contrasts using 10,000 random permutations of the RRPP (the observed case counting as one random occurrence). Observed effects and contrasts were considered significant if their P-values—estimated as percentiles in sampling distributions generated from RRPP—were less than an acceptable type I error rate of α = 0.05.

For every specimen, we employed two approaches to calculate the Bayesian posterior classification probabilities for assignment to subspecies: the first used equal prior probability of assignment to any subspecies, whereas the second used prior probabilities based on phylogenetic relatedness, as derived from phylogenetic covariances between subspecies, estimated from branch lengths. Prior probabilities were calculated as where c is the phylogenetic covariance between the j^th and k^th subspecies of the k = 9 subspecies, based on patristic distances between them assuming a Brownian motion model of evolution [44]. For any specimen, the k = 9 prior probabilities were calculated as such from the sum of all covariances between the actual subspecies designation, j, and each of the k hypothetical subspecies. Posterior probabilities were then calculated as where is a (transposed) vector of Procrustes residuals; k is the target group for association; is the multivariate normal density estimate, found as where d_M is the Mahalanobis distance between and the group k centroid, and C_W is the pooled within-group covariance matrix; is the normalization constant, solved as (i.e., the summation of all normal density estimates for all groups); and is the prior probability as previously described [45].

We calculated inter-quartile ranges (IQRs) of posterior probabilities for each subspecies, based upon specimen assignments. Morphological distinction for subspecies was identified from IQRs that did not overlap between true and hypothetical comparisons, such that specimens could be assigned to subspecies.

To evaluate dispersion of subspecific means, we first regressed Procrustes residuals against centroid size to yield allometry-free shape values. We used these data in a principal component analysis (PCA) to estimate ancestral characters states using squared-change parsimony [46–48]. This allowed us to map both phylogeny and subspecies shapes within a PC space so as to test the potential for adaptive radiation. The latter was accomplished by compiling shape dispersion relative to estimated ancestral states. We using a disparity through time (DTT) analysis [49] to test the dispersion of mean shapes for evidence of adaptive radiation. This involved plotting the relative morphological disparity (i.e., average disparity within clades divided by total disparity) versus “relative time” (i.e., a scale of 0–1 from root to most recent divergence, based on the distribution of tree nodes). We then compared within-subclade relative disparities versus that expected under a Brownian motion model of trait evolution, using 10,000 simulations to find the median disparity through time and its 95% confidence limits.

All analyses were performed in R, version 3.1.2 [50], with GPA and TPS calculated using GEOMORPH, version 2.1.3 [51–52]. Estimating ancestral character states, pruning of the phylogenetic tree, and assessing the chronogram were performed using APE, version 3.0–14 [53]. DTT analysis was performed using GEIGER, version 2.0.3 [54]. PC plots with ellipsoids for within-subspecies shape variation were generated using RGL (ver. 0.92.1098) [55].

Evolutionary history of the Western Rattlesnake

Despite two centuries of accumulating evidence, a comprehensive and formal reevaluation of Western Rattlesnake taxonomy has been slow to emerge. This reticence has been driven largely by three distinct, interrelated factors: (1) a long-standing disagreement over species concepts; (2) the convoluted history of this particular complex; and (3) an “essential tension” separating taxonomy and systematics. The first two are most relevant to our discussion and are parsed below.

By consensus, ‘species concepts’ represent the assumptions necessary for the recognition of morphological and genetic gaps among species [56, p. 447–472]. While these concepts do not delineate the gaps themselves, they provide a framework to interpret observed patterns [57–58], with the underlying premise that most, but not all, represent our attempts to explain or decipher how species came to be. This situation has been a source of confusion in that it conflates pattern with process. Our tact herein is to delineate gaps that correctly identify species rather than ponder mechanisms underlying the diversification.

The convoluted taxonomic history of the Western Rattlesnake complex is epitomized by the fact that no fewer than seven junior synonyms, 11 synonymizations, and 11 subspecific epithets have been applied to encompass its diversity [16, 59]. Fossil evidence for the complex is first recorded from the Late Pliocene [60] and extends through Late Pleistocene [61]. Molecular genetic data suggested a mid-Cenozoic origin, concomitant with the Neogene uplift of the North American Cordillera [62] presumably promoting the separation of eastern and western lineages, followed by constrictions and isolations within refugia during Pleistocene [16, 63] that drove additional differentiation within each. Shallow genetic divergence and relatively low diversity were manifested in subsequent radiations [16], resulting in species boundaries being obscured by incomplete lineage sorting [64] and ancestral polymorphism [65]. Both are artifacts prevalent in geologically recent and rapid diversifications, and they promote incongruent phylogenies and misinterpreted diversities when molecular data are used to derive taxonomic hypotheses.

Evolutionary divergence in the Western Rattlesnake

Our GM analyses employed dorsal and lateral configurations of the head, as well as semi-landmarks that capture shape via the use of outlines. Our results indicated that both size and subspecies designation are significant sources of head shape variation in the Western Rattlesnake complex. This, in turn, suggests that head shape has both ontogenetic and phylogenetic components, with the former being intuitively expected [66–67], whereas the latter was somewhat surprising, particularly given that head shape in snakes has been recorded as diagnostic only among genders [68–69], or within broadly based trophic guilds [70], some of which contained several species within a single genus [71].

To further dissect this relationship, we then evaluated all subspecies in pairwise comparisons (N = 36), and found significantly different head shapes for each, save the C. v. abyssus/ C. v. cerberus evaluation (P = 0.064; bold red, Table 2). We note that head shape variation has previously been recorded as diagnostic among species of a genus [72], yet without statistical significance. Such a diagnosis among individual subspecies, as recorded herein, is thus of considerable interest, and may stem from the combination of dorsal/ lateral/ and sliding landmarks, as employed in our GM analyses [72].

We also extended our shape analyses by applying a Bayesian Discriminant Function Analysis with uninform priors, and found that all subspecies reflected high posterior classification probabilities, with the exception of C. v. abyssus and its higher posterior probability of being assigned to either C. v. cerberus or C. v. lutosus. This is not surprising as they are sister taxa in the phylogeny (Fig 4). However, their head shapes are dissimilar with low posterior probabilities for the assignment of C. v. lutosus to C. v. abyssus (Q1 = 0.0015, Q3 = 0.0610). We interpret this as a reflection of the large 95% confidence limits that bracket C. v. abyssus, an aspect that adds an element of imprecision to individual assignments. In contrast, the head shapes of C. v. abyssus and C. v. cerberus are similar, and one might suggest homoplasy as a potential factor. However, the posterior probabilities for assigning C. v. cerberus to C. v. abyssus are again quite small (Q1 = 0.0025, Q3 = 0.0943), and this lack of reciprocity thus fails to support the assertion of homoplasy.

We also employed phylogenetically informed prior probabilities based upon an mtDNA phylogeny so as to improve our Bayesian posterior classification (Fig 6), and again noted a tendency for C. v. abyssus to assign with C. v. cerberus/C. v. lutosus. As before, posterior probabilities for the assignment of C. v. cerberus or C. v. lutosus to C. v. abyssus were again reduced, and neither could be correctly assigned to abyssus. We interpret this as suggesting (but not confirming) a distinct head shape in C. v. abyssus. While we recognize this may reflect instead its small sample size in our analyses, we also reiterate that its head shape is not homoplasious with either C. v. cerberus or C. v. lutosus.

Historic head shape and its contemporary divergence

Our depiction of evolutionary history in the C. viridis complex has focused rather intently on the contemporary morphospace of a mosaic structure, the head. Given this, we broadened and extended our analyses by (essentially) reconstructing an ancestral phenotype [as in 73] that potentially would have existed at each bifurcation of the tree (represented as numbered nodes in Fig 4). We accomplished this by employing squared change parsimony (a process that parallels maximum-likelihood but with a model of character evolution that invokes Brownian motion [74]) as a means to visualize an ancestral head shape). The process does not reconstruct outgroups or terminal taxa; those data were previously used in the estimation of ancestral character states. Rather, it acquires these comparisons by minimizing the sum of squared shape-changes along the branches of the phylogeny. This process assesses phenotypic divergences among terminal taxa as well as their estimated ancestral states. The assumptions and/or limitations in this process are several, but they are also tractable, such as: A constant rate of change (and a Brownian model of evolution); a correlation between neutral genetic and shape variation; and a deterioration in the reconstruction of character states as evolutionary time increases.

The results of these analyses indicated that pairwise shape distances among extant OTUs were significantly greater than those derived for ancestral character states, to include the ancestral node separating eastern and western lineages (Fig 4). In short, the closer the relationship between subspecies-pairs, the greater their differences in head shape. There are two mechanisms that can promote such a divergence among conspecifics: (a) competition, such that trophic character displacement is induced; or (b) a variance in foraging strategy that, in turn, promotes a shift in trophic ecology and, potentially, a dimorphism in size. Evidence from comparative analyses supports the latter [75], although its context relates to sexual dimorphism within and among snakes (and ramifications thereof), rather than to morphological divergence among groups in the broader sense (as herein). We explore this topic further in the section below.

Molecular/ morphological disparity and hypotheses

Evolutionary theory predicts a tight coupling between molecular divergence and morphological disparity, such that clades with large numbers of species (and thus elevated molecular variability) exhibit an equally broad morphological disparity [76]. Yet empirical studies do not bear this out [32,73], and in this sense, the Western Rattlesnake complex is another example of morphological disparity surpassing molecular genetic variation.

Specifically, we observed greater shape disparity between sister taxa (e.g. helleri + caliginis; lutosus + abyssus; viridis + nuntius) than we did for genetically less proximate pairings. Distances were significantly greater than expected under a Brownian model of evolution (Fig 5), and in turn, suggest an evolutionary force other than drift that underpins shape divergence. Interestingly, sister-pairs in the C. viridis complex are generally composed of a larger form and one more diminutive or stunted [77]. Thus, caliginis is recognized as a diminutive insular form of helleri, while abyssus is a stunted lutosus, and nuntius a dwarfed viridis. In each case, the geographic range of the stunted form is greatly reduced when compared to the larger sister taxon. Their trophic differences are also recognized [78] and provide empirical (but anecdotal) evidence in support of the argument that ecological competition rather than dispersal or environmental constraints is the primary factor that limits species distributions across large spatial and temporal scales [79].

Indeed, separation according to overall body size is readily achieved [80] in that size is minimally impacted by evolutionary stasis [81], and hence more easily malleable than overall shape, or shape-specific morphological structures. Body size drives the separation of sister-pairs over evolutionary time (as above), and diversifies species as they adaptively radiate [82]. However, rapid evolution is seldom observable and thus rarely detectable in fossil lineages [83], hence its recognition is always “after the fact.” We suggest that morphological displacement along a particular trajectory represents the residual of past competitive interactions and these are more easily detectable using GM approaches (as herein, [83–84]).

Our results demonstrate that morphological divergence between sister-pairs is greater in all cases than that derived overall. The head shapes of each OTU is distinct (save for abyssus), with the caveat that reciprocal assignment of one sister taxon to the other is not supported. We thus hypothesize that the significant disparity between sister-pairs in the C. viridis complex is both driven by body size and is resource-based. As such, divergence among OTUs represents the product of historical interactions, as recoverable in their morphologies.

Integrative taxonomy and the Western Rattlesnake

The C. viridis complex clearly requires taxonomic revision, yet this process has been problematic, with numerous attempts yielding either negative or incomplete results ([85] p. 569–570, [86]). Herein, we address this issue with a combined molecular and GM approach, as couched within a genotypic cluster criterion. And while we recognize the prescience of Klauber [87] in diagnosing nine OTUs to subspecies, we deem his approach as conservative in that it is based more on inferred intergradation than on quantitative differences. Klauber was trained as an electrical engineer and (we suggest) he refrained from diagnosing OTUs to species because he did not wish to counter the prevailing interpretation of a species-complex as being a ‘Rassenkreis’ (i.e., a ring of species, concomitant with expectations of intergradation among forms) [2,16]. Yet, at the same time, Klauber recognized ([18] p.164) that “…some of the newer methods of blood and venom studies may eventually indicate that the forms which we now consider viridis subspecies may really belong to two or more different species.” He seemingly realized the considerable weight of evidence differentiating these organisms, and understood that future studies might result in their re-classification as species.

We also recognize that a juxtaposition of legacy-based characters with more contemporary mtDNA, distributional, and GM data argues for a taxonomic reevaluation of the complex. Given this, we employed divergent data and methodologies so as to hypothesize species boundaries within a comprehensive and evolving framework termed ‘integrative taxonomy’ [88]. A review of integrative taxonomy identified 494 published studies over a seven-year period, with 233 of these (47%) using two different character types for discrimination [89]. Of these, 210 (90%) utilized DNA and morphology (as herein). Additionally, some 143 studies utilized vertebrates as study organisms, with 30% evaluating lepidosaurians (i.e., snakes, lizards, amphisbaenians, Tuatara), whereas another 30% assessed amphibians.

While an integrative approach has been influential in deriving species-boundaries in reptiles, its application has been multifaceted [88]. Some researchers suggest that concordance among unlinked molecular and morphological data is a requisite for species-delineation, in that it promotes taxonomic stability. Yet, one potential limitation is the risk of underestimating species numbers, with an assumption that characters evolve asynchronously within the speciation continuum. Others argue instead that integrative taxonomy should employ in its evaluation any available biological attribute, as long as those data provide evidence for the existence of a species. This, in turn, allows characters deemed most appropriate for a particular group to be utilized. Here, limitations stem from the application of a single line of discriminatory evidence (such as mtDNA), with an overestimation of species-numbers as a potential result. We elected to follow the first approach, with lineage divergence gauged across multiple characteristics and with the functional relevance of these data underscored in the speciation process [88].

Regardless, the application of integrative taxonomy necessitates a three-step protocol, as follows: (a) the accumulation of various data across numerous specimens; (b) the demarcation of groups using concepts that properly delimit species; and (c) the designation of nomenclature so as to recognize the new entities. However, one salient issue that emerged [90] was that a formal nomenclatural designation of a species failed to follow its character diagnosis. This discrepancy is rather disturbing for conservation, and thus we address this issue for the C. viridis complex in the following sections.

Taxonomic conclusions and recommendations

The integrative taxonomy that emerged from our comparison of mtDNA and geometric morphometric data was placed within the context of previous evaluations [16]. In so doing, we recognize six (of nine) C. viridis OTUs as phylogenetic species. We also supply a formal taxonomic nomenclature (Table 4) that defines their separate evolutionary trajectories, with additional amplification (below) regarding those OTUs subsumed into one of the six designated species:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. A revised taxonomy for six species of the Crotalus viridis complex based upon molecular and morphological data and an integrated taxonomy approach.

Standard English names follow [1,7].

https://doi.org/10.1371/journal.pone.0146166.t004

Our combined data and that of others validate two highly divergent lineages of C. viridis (eastern and western, respectively). The eastern group is recognized as C. viridis, but with C. v. nuntius subsumed within C. viridis (as before). However, we note that multiple lineages within the western clade reflect high mtDNA sequence divergence [16] as well as significant shape divergence (our data). Furthermore, these OTUs occur in relatively discrete geographic regions and seemingly reflect little intergradation or hybridization. Given this, and following a contemporary and theoretically well-grounded approach to species delineation (as above), we recommend the elevation of cerberus, concolor, helleri, lutosus, and oreganus to full species status, as based upon their reciprocal distinctiveness across both mtDNA and GM data.

The insular distribution of C. v. caliginis has previously been suggested as a condition sufficient for specific status [91], and our GM data support this as well. However, a close association between calignis and helleri, at the mtDNA level [14–16] suggests instead calignis is not sufficiently divergent at the molecular level and should instead be designated as an insular population of C. helleri. The situation is similar with C. v. abyssus, but with discriminating data in the integrative taxonomy essentially reversed. In this sense, abyssus was previously suggested as worthy of species-recognition based upon mtDNA data [16], yet the GM data from the current analysis (S1 Fig, Fig 6) fail to provide a statistically significant (and thus corroborative) level of support. Given our combined and integrative approach fails to differentiate abyssus, we recommend it should instead be subsumed within C. lutosus.

Implications for conservation and management

Species, subspecies, and ‘distinct population segments’ of vertebrates can be listed under the ESA [2,92], albeit with clear requisites. For example, the extent of scientific uncertainty associated with the listing must be explicitly recognized, such that the conservation decisions that ensue can be more readily advocated and accepted [93]. These issues have proven problematic for polytypic species complexes whose taxonomies have most often been qualitatively described from reduced numbers of individuals sampled ambiguously across broad geographic continua. These derivations conflicted with evolutionary and statistical methodologies and were thus inappropriate for conservation planning and the phylogenetic perspectives it required. Their recognition consequently deteriorated under more contemporary scrutiny, as did the potential listing proposals based on their older taxonomies [89]. This has been especially true for herpetological subspecies that are numerically rich but with numbers that have decreased ~50% over the last decade [89].

It is important to recognize the numerous benefits to habitat conservation that have accrued from such a reorientation, with management units (MUs) [89,95], evolutionarily significant units (ESUs) [92,94,95], or even cryptic but distinct species [2, herein] being defined by application of more contemporary methodologies. As a result, connectivity can be promoted, fragmentation minimized, and the conservation value of the habitat subsequently elevated if indeed such patterns represent composites across multiple species [63]. In this sense, the adaptive management of multiple taxa should promote, or at least sustain, the geographic integrity of a region [2].

Our considerations underscore the influence of taxonomy on conservation, but with several recognized imperatives: the necessity of a regional- rather than taxon-centric view of management, and the promotion of a taxonomy that is contemporary, well defined, and represents the best science available.