The evolution of tail length in snakes associated with different gravitational environments

Tail Length Data

Data for tail length were collected from the literature, museum specimens and live snakes representing 226 species within 139 genera and 15 families (Uetz & Hosek 2013; see Table S1, Supporting Information). H. B. Lillywhite and R. S. Seymour obtained additional unpublished data for snake length opportunistically during previous investigations into the anatomy and physiology of snakes. In combination, these data represent almost all snake families (Uetz & Hosek 2013). Museum specimens were acquired from the Florida Museum of Natural History (FLMNH) and the National Museum of Natural History, Smithsonian (USNM). Only adult females were used for this study to avoid the potentially confounding effects of sexual dimorphism in RTL and ontogenetic shifts in allometry and habitat use (Klauber 1943; King 1989; Shine 2001). A snake was determined to be an adult if its TL was within, or very near, the published size range for adults of that species. For museum specimens, sex was determined by subcaudal incision. Live snakes were measured at Glades Herp, Inc., Florida, and were probed to determine sex. Taxa in which females are known to possess well-developed hemipenis-like structures (e.g. Pseudoficimia and Gyalopium; Hardy 1972; Smith & Brodie 1982) were not included in this study. Some Bothrops insularis females have hemipenis-like structures (Hoge et al. 1961), and this species was included in the study under the assumption that length data reported by Klauber (1943) were from functional females.

Snout-vent length (SVL) was measured from the tip of the snout to the posterior edge of the anal scale. Tail length was measured from the posterior edge of the anal scale to the tip of the tail, and only snakes with complete tails were used in this study. SVL and tail length were measured using either a meter stick (±1·0 mm) or a calliper (±0·1 mm). A string was used to follow the body contours of live snakes and rigid specimens and was subsequently measured with a meter stick. Length measurements were repeated on individuals up to 10 times when possible and then averaged. We primarily used mean SVL and tail lengths from the literature or from multiple specimens, but in some cases, we used data representing single specimens. To test intraspecific variation in RTL, we collated data from the literature on the total variation in adult female RTL for 26 ecologically diverse snake species in five families and 26 genera [mean sample size per species = 58·58 ± 11·94 (SE), range = 5–249] (Klauber 1943; Broadley 1959). Mean intraspecific variation in RTL (±SE) was extremely small (1·0% ± 0·003), allowing us to justify the use of RTL data for individual specimens when necessary.

Constructing a Snake Phylogeny

A consensus phylogeny of extant snake species does not currently exist. Therefore, we constructed a composite tree in order to examine our data within a phylogenetic context. Phylogenies constructed by Wiens et al. (2008) and Pyron et al. (2011) were combined manually in MacClade version 4.08 (Maddison & Maddison 2005) and used as a base phylogeny, to which additional species or clades were added or expanded following the methodology outlined by Lillywhite et al. (2012). Phylogenetic relationships for five genera were unresolved (i.e. Atractus, Dipsas, Langaha, Philothamnus and Prosymna). With a single exception (Philothamnus hoplogastor), these polytomies contain congeners that are categorized within the same gravitational habitat, and results of phylogenetic analyses were not affected when polytomies were resolved using all possible alternative topologies (results not shown). The composite tree is overall consistent with other recent molecular analyses of snake relationships (e.g. Noonan & Chippindale 2006; Vidal et al. 2010; Pyron et al. 2011, 2013; Pyron & Burbrink 2012; Wiens et al. 2012; Pyron, Burbrink & Wiens 2013) regarding the relative placement of higher level taxonomic groups (i.e. Scolecophidia, Alethinophidia and Colubroidea).

The following phylogenetic studies were used to add taxa to the base phylogeny to construct the composite tree (Fig. 1). The scolecophidian clade was replaced using the phylogeny of Adalsteinsson et al. (2009). The phylogeny of Grazziotin et al. (2012) was used for placement of taxa within the subfamilies Dipsadinae and Xenodontinae. Relationships among the sea snake genera Aipysurus, Emydocephalus and Hydrophis were inferred using the phylogeny of Sanders et al. (2013).

Categorizing Gravitational Habitats

Because snakes occupy a wide variety of habitat types, it is often useful to divide habitat usage into generalized categories when discussing snake community assemblages. Perhaps the most widely utilized categories for snake habitat use include fossorial, terrestrial, aquatic, semiaquatic, arboreal and semiarboreal, or some subset of these (e.g. Johnson 1955; Shine 1983; Guyer & Donnelly 1990; Dalrymple et al. 1991; Lindell 1994; Vidal et al. 2000).

However, as noted by Johnson (1955), habitat categories per se can be inadequate when comparing ecomorphological or behavioural data between species. For example, watersnakes in the genus Nerodia typically live and obtain food near water (Conant & Collins 1998). Consequently, they are usually categorized as aquatic or semiaquatic (e.g. Vidal et al. 2000). However, during parts of the year, some Nerodia species can spend as much or more time above the ground in shrubs and trees than Coluber constrictor, a snake typically categorized as semiarboreal (Mushinsky, Hebrard & Walley 1980; Plummer & Congdon 1994). Separating Nerodia and Coluber into distinct ecological categories (aquatic or semiaquatic versus semiarboreal) obscures these similarities in habitat usage. By the same token, categorizing the partially aquatic and partially terrestrial Nerodia and entirely pelagic Hydrophis platurus as ‘aquatic’ is misleading because it does not account for the variation in habitats encountered by the former. To avoid these problems, we employed both habitat use and behavioural information when assigning snake species into gravitational habitat categories (Lillywhite et al. 2012). These gravitational habitat (G-habitat) categories reflect the inferred (maximum) gravitational stress that snakes are likely to experience while utilizing various habitats, which is correlated with the frequency with which snakes assume upright or vertical postures, and especially fully vertical positions (Lillywhite et al. 2012).

In order to categorize the amount of inferred gravitational stress snakes experience, we created three broad categories of gravitational habitat: (i) stenotopically (narrowly adapted) arboreal, (ii) eurytopically (broadly adapted) arboreal/terrestrial and (iii) non-scansorial (Lillywhite et al. 2012). Stenotopically arboreal species primarily live in arboreal habitats. These species occupy trees or vegetation >50% of the time, and they are presumed to experience the greatest degree of gravitational stress. Eurytopically arboreal/terrestrial species are often found on the ground, but they regularly climb for reasons including foraging, escaping predators and thermoregulation (e.g. Masticophis and Coluber). The non-scansorial category comprises both ground-dwelling terrestrial and aquatic snake species that rarely if ever climb, and these species are presumed to experience the least degree of gravitational stress of all three G-habitat categories. The RTL data for terrestrial and aquatic species were pooled for analyses because they were not statistically different (ancova: P = 0·969). However, we discuss aquatic and terrestrial species separately within the context of cardiovascular adaptation to gravitational stress. Some characteristically terrestrial species, or populations of species, do climb occasionally (e.g. Bitis arietans, Bitis armata, Bothrops asper, Crotalus horridus and Thamnophis sirtalis) (Sajdak 2010). However, we consider this behaviour atypical and do not consider them eurytopically arboreal/terrestrial. Aquatic species are those primarily found in aquatic habitats and which commonly exhibit one or more morphological specializations functionally associated with aquatic habitats such as a laterally compressed or oar-like tail, more dorsally positioned eyes and nostrils, salt excreting glands and valvular nostrils (Heatwole 1999). Examples include laticaudines (genus Laticauda) and the homalopsine Enhydris enhydris, which are clearly adapted to an aquatic lifestyle and are usually found in water, even though they may occasionally sojourn onto land. Aquatic species are presumed to experience the least gravitational stress because the surrounding water column acts as an ‘antigravity suit’ (Lillywhite 1993a: 561). Importantly, all categorical placements were based entirely on the ecology and behaviour of each species and not on morphology. These gravitational habitat categories have been proposed to represent a variable that is a continuous trait in nature (Lillywhite & Henderson 2001).

Categories of gravitational habitat were based on adult uses of habitat compiled from the literature and from the numerous observations in nature made by others and ourselves. A large amount of this information were obtained from the following sources: Africa (Broadley & Cock 1975; Broadley 1983; Branch 1998; Schmidt & Noble 1998; Spawls et al. 2002); Australia (Shine 1995); Central America (Campbell 1998; Savage 2002); India (Whitaker & Captain 2004); Madagascar (Glaw & Vences 1994; Henkel & Schmidt 2000); North America (Smith & Brodie 1982; Conant & Collins 1998); South America (Murphy 1997; Boos 2001; Duellman 2005); and the West Indies (Schwartz & Henderson 1991).

Statistical Analyses

We used conventional regression analyses, analysis of covariance (ancova), analysis of variance (anova) and phylogenetically independent contrast (PIC) analyses to quantify relationships between snake morphology and gravitational habitat. We tested for interspecific differences in RTL and relative snout-vent length (RSVL) using ancova with tail length or SVL as the dependent variable, G-habitat as the factor and TL as a covariate. The results of the ancova were not affected when tail length was used as the dependent variable, G-habitat as the factor and SVL was instead used as a covariate. Thus, we only report ancova results using TL as a covariate since it is more biologically relevant to our central hypothesis. We report ancova results as adjusted means, which are the predicted values of the dependent variable (tail length or SVL) for each G-habitat where the covariate variable TL is evaluated at the average of its data values. Residuals from linear regressions are often used for the purpose of controlling for unwanted effects in multivariable data sets. However, several studies have identified ancova as a more appropriate statistical method to control for covariates (García-Berthou 2001; Freckleton 2002). Coefficients of variation (CV) were calculated to compare variation in TLs among different G-habitats. Data were log₁₀-transformed prior to statistical analyses. Conventional statistical analyses were performed using the statistical program package sigmaplot version 13 (Systat Software, Inc., San Jose, CA, USA). Data are reported as mean ± SE unless stated otherwise.

Interspecific comparative analyses can be confounded by pseudoreplication caused by phylogenetic relatedness among species samples (Price 1997). However, these effects can be partially resolved using PICs (Felsenstein 1985), which is a widely used method for phylogenetic correction (e.g. Pizzatto, Almeida-Santos & Shine 2007; White, Blackburn & Seymour 2009; Lillywhite et al. 2012; Adams 2014).

Analyses of PIC between tail length (expressed as a decimal fraction of the TL) and G-habitat were performed on log₁₀-transformed data using the pdap: pdtree v. 1.74 module of mesquite version 2.75 (Maddison & Maddison 2011). Branch lengths could not be estimated because the final phylogenetic tree was a composite from a variety of sources. Independent contrasts were therefore generated with branch lengths assigned as either equal or arbitrary in Mesquite using ‘Branches Proportional to Lengths’, with each branch having a length of 1·0, under the assumption that clade age is proportional to number of species (Grafen's branch lengths; Grafen 1989; Maddison & Maddison 2005). Plots of standardized contrasts against the variance of untransformed contrasts showed strongly significant correlations for Grafen's, but not equal or arbitrary, branch lengths. As significant correlations between branch lengths and trait values violate a key assumption of independent contrasts analysis (Diaz-Uriarte & Garland 1996), we elected to use equal branch lengths. Contrasts were calculated between all sister lineages on the tree for G-habitat and RTL, and relationships were examined between the variables by calculating regressions on these standardized contrasts using least squares (Garland, Harvey & Ives 1992; Grafen 1992). Absolute values of contrasts were not correlated with their standard deviation, alleviating the need for branch length transformations (Diaz-Uriarte & Garland 1998).

The three gravitational habitats (i.e. non-scansorial, eurytopically arboreal/terrestrial and stenotopically arboreal) were analysed in the phylogenetic independent contrast analysis under the reasonable assumption that the degree of gravitational stress experienced forms a continuum with stenotopically non-scansorial < eurytopically arboreal/terrestrial < stenotopically arboreal. Operationally, any discrete trait value can be treated as continuous in mesquite 2.75, and this arrangement represents a more realistic assumption than gravitational habitats assigned as discrete states (Pizzatto, Almeida-Santos & Shine 2007; Lillywhite et al. 2012). To more directly compare the results of non-phylogenetic and phylogenetic analyses, we also performed a non-phylogenetic linear regression of RTL (as a decimal fraction of TL) against G-habitat treated as a continuous variable.

Body Morphology, Habitat and Gravity Stress

These results can be explained within the context of adaptive response to cardiovascular stresses induced by gravity while climbing. Lengthening of the tail relative to body size increases the relative segment of elongate blood vessels that is surrounded by tight (low compliance) tissues. The subcutaneous compliance of caudal tissue in arboreal snake species is comparatively low and increases up to sixfold in non-scansorial species (Lillywhite 1993a). Moreover, postural shifts of fluid in arboreal snake species are 1/3 or less than those in aquatic snake species that are unaffected by gravity in their neutrally buoyant habitat (Lillywhite 1985). The low caudal compliance of arboreal snake species helps mitigate posterior blood pooling, thereby facilitating the return of blood flow to the heart. This in turn increases cardiac outflow to the brain and vital organs during upright positions that are experienced during climbing (Lillywhite & Gallagher 1985; Lillywhite & Donald 1994; Lillywhite 2005).

Although we found that arboreal snakes in general have longer RTLs (and therefore shorter RSVLs) than non-scansorial species, the results of the ancova suggest that selection on RTLs may be more finely graded than previously thought, given that the most highly arboreal (stenotopically arboreal) species have longer RTLs than eurytopically arboreal/terrestrial species. Previous studies have grouped stenotopically arboreal and eurytopically arboreal/terrestrial species into a single (scansorial) gravitational habitat (e.g. Lillywhite et al. 2012). The results reported herein suggest that future studies investigating adaptive responses to gravitational stress in snakes, and possibly in other vertebrates, can consider finer degrees of cardiovascular stress when evaluating the effects of gravitational habitat.

Regressions of log tail length and log SVL, with log TL as a covariate, for the three G-habitats had different y-intercepts but slopes that were not significantly different (Figs 3 and 4, respectively). These findings suggest that stenotopically arboreal species have longer tails and shorter bodies (SVL) than eurytopically arboreal/terrestrial species at all TLs. Furthermore, eurytopically arboreal/terrestrial species have longer tails and shorter bodies than non-scansorial species at all TLs.

These results are readily explained in the context of physiological trade-offs in response to gravity stress. Upright postures create vertical gradients of gravitational pressures within circulatory vessels, and the magnitude of the pressure change is proportional to the total length of the blood column. In air, this potentially induces blood pooling and oedema in dependent tissues and possibly decreases the volume of blood reaching the head and vital organs of animals that are tall or elongate. Thus, snakes with longer TLs experience larger gradients of gravitational pressure than do shorter snakes. The negative effects of body length on blood circulation are partly offset in arboreal/scansorial species by having longer tails relative to SVL, resulting in a greater proportion of TL that is characterized by lower compliance tissues (Lillywhite 1985, 1993a). Thus, stenotopically arboreal species with long TLs have even longer RTLs than eurytopically arboreal/terrestrial species. These results support the hypothesis that phylogenetic patterns of tail length represent adaptive responses for maintaining systemic blood pressure (and cephalic blood flow) while climbing.

Our results also reaffirm previous conclusions regarding the use of TL, not SVL, as the most appropriate variable to use when investigating cardiovascular adaptations in snakes, because TL more accurately measures the column-length of the circulatory system (Lillywhite & Seymour 2011; Lillywhite et al. 2012). SVL is often useful for investigating various ecological and evolutionary questions in snakes, largely because it appears to better predict body mass than does TL (Feldman & Meiri 2013). However, with regard to investigating evolutionary adaptations of the cardiovascular system to gravitational stress, the entire fluid column (i.e. TL) is under selection, not just the vessels in the pre-vent portion of the body (SVL). Thus, omitting tail length from such studies omits varying percentages of the total fluid column upon which natural selection is acting.

Tail Length and Functional Ecology: Other Considerations

Although RTL data for aquatic and terrestrial species are pooled into a ‘non-scansorial’ G-habitat category for statistical analyses, the effects of gravity on blood circulation in these species are different. The range of tail lengths as a percentage of TL for the aquatic species (9·3–19·2%) falls entirely within the range for the terrestrial species (1·1–31·1%). The vascular blood columns of aquatic snakes experience little or no gravitational stresses associated with an essentially weightless environment (Seymour & Lillywhite 1976; Lillywhite & Pough 1983; Lillywhite 1996). Therefore, in aquatic environments, tails do not function in countering the effects of gravity on blood circulation, and tail morphology is more constrained by forces of propulsion and locomotion. Consequently, aquatic species are expected to have tail lengths long enough to effectively propel the snake through water, yet short enough to minimize drag (Aubret & Shine 2008). Although the role of tail length in snake locomotion remains incompletely documented (e.g. Jayne & Bennett 1989), similarities in tail lengths between aquatic and terrestrial species are here hypothesized to result from different selective pressures.

Mean snout-vent length relative to total body length (RSVL) shows a relationship opposite to that of RTL, decreasing with increasing use of arboreal habitat among the three G-habitat categories of the present study (Fig. 2b). However, mean TL was not significantly different between stenotopically arboreal and eurytopically arboreal/terrestrial species, although both categories had longer mean TLs than non-scansorial species (Fig. 2c). Thus, TL appears to be constrained in (adult female) stenotopically arboreal snakes. Macrohabitat use imposes locomotory constraints upon snakes (Jayne & Herrmann 2011), and there is considerable evidence that arboreal snakes are limited in their use of habitat as a consequence of interactions between their morphology and the physical size of branches (Lillywhite & Henderson 2001; Astley & Jayne 2007; Jayne & Herrmann 2011). Arboreal locomotion appears to primarily involve various combinations of undulatory and concertina movements along a complex and often unstable three-dimensional substrate (Gans 1974; Edwards 1985; Lillywhite & Henderson 2001; Astley & Jayne 2009). In addition to the cardiovascular constraints associated with long TL, large-bodied or heavy-bodied snakes may not be adequately supported by smaller branches, and they may have more difficulty spanning gaps (Henderson & Nickerson 1976; Jayne & Herrmann 2011; Ray 2012). Indeed, the relationship between gap-spanning ability and body length shows a negative allometry in one stenotopically arboreal species (Boiga irregularis), indicating that smaller individuals are able to span gaps that are a larger percentage of their length than are larger individuals (Lillywhite et al. 2000; Jayne & Riley 2007). Thus, due to physiological limitations and the requirement of foraging snakes to span gaps between stems and branches, it is reasonable to expect strong selection on the upper limits of TL in arboreal snake species. Furthermore, among the three G-habitats of this study, stenotopically arboreal species had the lowest CV for log TL, non-scansorial species had the highest variation, and values for eurytopically arboreal/terrestrial species were intermediate. These patterns of diversity are exactly what would be predicted if the use of arboreal habitat constrains the maximum TL.

Some snake species demonstrate ontogenetic shifts in macrohabitat use, with juveniles being found more frequently utilizing arboreal macrohabitats than adults (Martins & Oliveira 1999). This behavioural shift has been well documented in several large boid species such as Python sebae (Spawls et al. 2002) and Boa constrictor (Campbell 1998), some long colubrids such as Pantherophis alleghaniensis (H. Lillywhite pers. obs.) and Pseustes poecilonotus (Boos 2001), and some very long elapids such as King Cobras, Ophiophagus hannah (R. Whitaker pers. comm.). Although there are many potential reasons for ontogenetic shifts in macrohabitat use to occur in snake species that attain long total lengths as adults, this pattern supports the hypothesis that there are constraints on the total length of arboreal snakes.

The 25 viperid snake species included in this study exhibited relatively short tail lengths, and the arboreal viperids exhibited relatively short total lengths (see Table S1). The Viperidae is an interesting group because a significant number of arboreal specialists have evolved within the clade, but these are relatively short in body length. The shorter length minimizes the gravity issue, so other adaptations typically seen with respect to cardiovascular specializations in arboreal species are not required in the shorter snakes (Lillywhite et al. 2012).

Many arboreal snake species have prehensile tails, and some species use the tail to anchor to branches in order to generate forces necessary to bridge gaps (Jayne & Riley 2007). However, many arboreal snakes are also able to grip branches using the body (Jayne & Herrmann 2011). Tail prehensility does not appear to be correlated with the longest tail lengths (Lillywhite & Henderson 2001). This is perhaps because, with sufficient force, the tails of a number of terrestrial and some eurytopically arboreal/terrestrial species with long tails may break as a defensive mechanism (Greene 1988; Guyer & Donnelly 1990). However, studies investigating the correlation between tail length and tail prehensility in snakes have looked at snakes in general. RTL may be correlated with tail prehensility in some clades more than others, but this hypothesis remains to be tested. Nonetheless, the extent to which arboreal snakes use their tails for locomotion remains poorly studied (Jayne & Herrmann 2011). In terrestrial environments, there seems to be no compelling reason why increasing the length of tail should confer advantages to locomotion in arboreal vs. strictly ground-dwelling environments. Evidence suggests that relatively long tails are not functionally advantageous for rapid terrestrial locomotion in snakes (Jayne 1988; Jayne & Bennett 1989).

Tail Length and Fecundity

The inverse relationship between RSVL and inferred gravitational stress is surprising given the generally intense selection for fecundity and a longer SVL in female snakes (Madsen & Shine 1992; Shine 2001). Body size (SVL) and litter or clutch size are often positively correlated in snakes (Vitt & Vangilder 1983; Madsen & Shine 1992; Seigel & Ford 2001), including some stenotopically arboreal species (Scartozzoni, Salomão & de Almeida-Santos 2009). However, the pattern of decreasing RSVL with increasing gravitational stress suggests there is an evolutionary trade-off between relatively long tails and fecundity in arboreal species. The trade-off appears to favour longer tail lengths (or shorter RSVL) over fecundity as arboreal habitat-use increases. In species of arboreal snakes for which data are available, most have relatively small clutches or litters (e.g. Martins & Oliveira 1999; Pizzatto, Almeida-Santos & Shine 2007; Hedges 2008), and this has been attributed in large part to their relatively thin bodies (Lillywhite & Henderson 2001; Sajdak 2010). Shorter SVLs would further constrain fecundity in female arboreal snakes by greatly reducing the body cavity space available for eggs or young. Some arboreal snake species may have evolved morphological and life-history traits to ameliorate fecundity constraints such as paired ovaries having minimal or no spatial overlap along the length of the body cavity (Pizzatto, Almeida-Santos & Shine 2007), and possibly the production of multiple clutches or litters per year (Sajdak 2010) along with differences in egg size and shape (Hedges 2008). Nonetheless, arboreal species appear to persist in relatively high densities in some snake communities (e.g. Guyer & Donnelly 1990; Duellman 2005), suggesting that selection for fecundity and associated life-history traits in arboreal species provide a potentially fruitful area of investigation.