Museum für Naturkunde—Leibniz-Institut für Evolutions—und Biodiversitätsforschung Invalidenstr. 43, Berlin, Germany
Institut für Biologie, Humboldt-Universität zu Berlin, Berlin, Germany
Berlin-Brandenburg Institute of Advanced Biodiversity Research (BBIB), Berlin, Germany
Correspondence to: Johannes Müller, Museum für Naturkunde—Leibniz-Institut für Evolutions—und Biodiversitätsforschung Invalidenstr. 43, D-10115 Berlin, Germany. Fax: +49-30-2093-8868. E-mail: firstname.lastname@example.org
Amphisbaenia, or “worm lizards,” is an enigmatic, poorly understood clade of fossorial, body-elongated squamates, nearly all of which are limbless. Although they have historically been associated with snakes (e.g., Rieppel, 1988; Conrad, 2008) or various lizards (e.g., teiids; Boulenger, 1884), molecular and paleontological evidence now supports the position of amphisbaenians as sister group to Lacertidae, an Old-World clade of fully limbed, surface-dwelling lizards (Townsend et al., 2004; Vidal and Hedges, 2005; Muüller et al., 2011). Within Amphisbaenia, current systematics based on DNA and shared derived morphology support the recognition of five distinct lineages: the diverse Amphisbaenidae, and the species-poor Trogonophidae, Blanidae, Bipedidae, and Rhineuridae (Kearney, 2003; Kearney and Stuart, 2004; Macey et al., 2004; Vidal et al., 2008). Relationships within Amphisbaenia remain uncertain, with morphological and molecular topologies in strong contradiction. The morphological analysis of Kearney (2003) showed the Bipedidae, the only clade possessing (fore)limbs, as most basal and sister to all other amphisbaenians. In contrast, molecular-based phylogenies place the North American Rhineuridae at the base of the tree (Kearney and Stuart, 2004; Macey et al., 2004; Vidal et al., 2008; Wiens et al., 2012), implying that limb loss occurred multiple times in amphisbaenian history.
Most extant amphisbaenian families, with the exceptions of Amphisbaenidae and Trogonophidae, contain only a single genus restricted to a small geographical area. This is most extreme in Rhineuridae, whose modern distribution is a classical example of relict endemism. Rhineurids are currently represented by a single species, Rhineura floridana, restricted to the Florida Peninsula (Mulvaney et al., 2005), despite a rich fossil record for the Cenozoic of North America (Estes, 1983; Hembree, 2007). Rhineurids were present during large parts of the Paleogene and early Neogene in the western United States, but appear exclusively in Florida after the mid-Miocene (Estes, 1983). Their fossils are mainly known from a number of well-preserved skulls that appear nearly identical to the modern form, characterized by a “shovel-headed” snout with a steep craniofacial angle and dorsoventral compression. This specialized anatomy is associated with the fossorial habits of R. floridana and other amphisbaenians, which use their heads as digging tools in loose or sandy soils (Gans, 1969, 1974).
The high degree of morphological conservatism among rhineurids over time suggests strong evolutionary constraints on phenotypic variation. Such constraints may limit the ability of organisms to adapt to environments outside of their ancestral ranges, resulting in restricted geographical distributions (Wiens and Graham, 2005). In this context, the relictual distribution of R. floridana may be the result of stabilizing selection on ecological traits related to fossoriality, such as head shape. Instead of adapting to new habitats as climates, and potentially soils, changed in North America, rhineurids may have shifted their geographical ranges to remain within the ancestral niche. The surviving taxa would therefore become restricted to the only area retaining presumably favorable conditions, the humid subtropical regions of Florida.
The tendency of lineages to retain ancestral niche-related traits, and thus geographical ranges, is known as phylogenetic niche conservatism (PNC) (Wiens et al., 2010). Tests for PNC generally involve trait comparisons among closely related taxa, with the requirement that they be more similar ecologically than would be expected based on phylogenetic relationships alone (Losos, 2008). Detecting PNC in clades containing only a single species is therefore problematic, particularly when the closest living relatives, as those of rhineurids, show radically different environmental preferences (i.e., primarily semi-arid/Mediterranean affinities for blanids, bipedids, trogonophids, and many amphisbaenids). In such instances, fossils may yield further insights into a clade's ancestral ecology, which can be used to test for the presence, or at least the extent, of PNC. However, surprisingly few studies have considered the environmental conditions of an extant taxon's fossil relatives as direct indicators of ancestral niche preferences (e.g., Royer et al., 2003), thus limiting our ability to detect PNC over broader taxonomic and temporal scales.
In this study, we apply an integrative approach incorporating paleontological and neontological information to test if PNC contributes to modern R. floridana's restricted distribution. We test for PNC in extant and fossil Rhineuridae using a two-scaled approach in which we: (1) characterize environmental preferences of extant R. floridana using species niche modeling for comparisons with fossil R. floridana, and (2) reconstruct paleoenvironmental tolerances of extinct Rhineuridae based on occurrences of fossil rhineurids across the Cenozoic of North America. This approach allows us to test if modern R. floridana is ecologically conservative, which may explain its currently restricted distribution, and if PNC is characteristic of Rhineuridae as a whole. A potential shortcoming of this approach is that we are limited to abiotic factors such as temperature and precipitation, although we recognize that biotic factors, including competition, food availability, and predation, may also play important roles in species distributions.
MATERIALS AND METHODS
Species Niche Modeling
To compare climatic factors characterizing fossil and extant rhineurid distributions, we constructed an environmental niche model for modern R. floridana in MaxEnt version 3.3.3k (Phillips et al., 2006). MaxEnt estimates species distributions by minimizing the distance between probability densities of presence records and environmental variables in covariate space (Elith et al., 2011). In the absence of information on species prevalence, MaxEnt uses a prevalence parameter τ to describe the probability of presence at sites with “typical” environmental conditions for the species, that is, close to the multivariate mean extracted from the presence data. MaxEnt has been shown to perform well with small sample sizes and for species with small geographic ranges (Hernandez et al., 2006), making it appropriate for identifying environmental factors associated with R. floridana's current distribution.
Georeferenced records for R. floridana were obtained from various sources: the Florida Museum of Natural History, Georgia Department of Natural Resources, Mulvaney et al. (2005), the HerpNET data portal, and the Global Biodiversity Information Facility. After controlling for quality (i.e., terrestrial records within the reported range), 303 records were available for model construction. For environmental variables, we focused on characteristics of temperature, precipitation, and soil that could be directly compared with information derived from the fossil record. Eleven bioclimatic variables (BIO1–7, 12–15) reflecting annual trends, seasonality, and extremes of temperature and rainfall were downloaded from the WORLDCLIM database at 30 arc-sec (∼1 km2) resolution (Hijmans et al., 2005). Spatial information on dominant soil type, composition, and texture was taken from the Harmonized World Soil Database v.1.2 (FAO/IIASA/ISRIC/ISSCAS/JRC, 2012). Both the soil layer and bioclimatic variables were incorporated into the MaxEnt model.
Ten replicates of the niche model were run for 5,000 iterations each to allow for adequate convergence among runs. Cross-validation sampling was used to test the model's predictive power. As opposed to subsampling, this method uses random equal-size partitions of the entire data set and is therefore better suited for small numbers of occurrences (Elith et al., 2011). The averaged prediction over all runs was used in generating the final niche model. We used the logistic model output with a default prevalence of 0.5, essentially bounding the probability of presence between unsuitable and suitable areas between 0 and 1 (Elith et al., 2011). Locations (i.e., cells) with a logistic output of 0.5 represent a 50% probability of presence of conditions considered suitable to the species' survival, and hence a 50% chance of the species being present.
Phylogeny of Extant and Fossil Rhineuridae
Due to the discrepancy between morphological and molecular investigations of amphisbaenian relationships (see above), we performed a total evidence analysis incorporating information from both sources. For morphological data, we used the character matrix of Kearney (2003), consisting of 162 characters describing external anatomy, dentition and osteology. Some potential rhineurid fossils were excluded from the phylogenetic analysis due to poor preservation or insufficient character information (e.g., early Miocene “Rhineura marslandensis” and “R. sepultura”). Only rhineurid fossils with diagnostic material were included, plus two fossil skulls from the Miocene of Africa (Lophocranion rusingensis and Listromycter leakeyi; Charig and Gans, 1990). Lacertid lizards (scored as a single taxon) were used as an outgroup. We also added the recently described stem-amphisbaenian Cryptolacerta hassiaca from the Eocene of Germany (Muüller et al., 2011).
The morphological data set was combined with two nuclear genes: 2,842 base pairs (bp) of RAG1 and 374 bp of CMOS from Müller et al. (2011; see information therein for Genbank accession numbers). Because gene sequences were not available for many individual amphisbaenian species, we merged all amphisbaenids, blanids, bipedids, and trogonophids into single taxa, using the morphological data of Amphisbaena alba, Blanus cinereus, Bipes biporus, and Trogonophis wiegmanni as representatives of their respective clades. Gene sequences of Eremias arguta were used for Lacertidae. The full combined morphological and molecular matrix is available on Morphobank (http://morphobank.org/permalink/?P974).
Bayesian phylogenetic analysis of the combined data sets was performed in Mr Bayes 3.2.1 (Huelsenbeck and Ronquist, 2005). Two parallel runs with four chains each were run for 2 million generations, with sampling every 100th generation and a burnin of 5,000. The average standard deviation of split frequencies was 0.005126, indicating that the two runs sufficiently converged.
Paleoclimate of Rhineuridae
We used global, regional and local paleoclimate data to infer environmental conditions of extinct rhineurids. In Table 1, we summarize paleoclimate for each fossil taxon based on independent studies of isotopes, paleobotany, and paleosols regarding its respective geological formation and age. We also illustrate these changes on our generated rhineurid phylogeny, to show the timing of paleoclimatic events in relation to rhineurid diversification.
Table 1. Listing of all currently valid taxa of extant and fossil Rhineuridae, with age, fossil localities, and the inferred paleoclimate and environment
Not much known, but the Illinoian is generally considered a glacial period; however, Holman (1959a) notes the striking similarity of the herpetofauna to what is known from there today. Holman (1959b) concludes that the site was probably marshy pineland and that the climate was a bit cooler than today as indicated by mammal fossils
Oligocene drying trend continues, stream channel sand and floodplain mud (paleosols) typical of steppe or even desert-like conditions (Stoffer, 2003), open grassland with annual rainfall of 350–450 mm (Prothero, 1994)
Seasonal drying, potentially reaching freezing temperatures in winter, open deciduous and coniferous forest similar to modern southern Chinese highlands (Dillhoff et al., 2009), annual rainfall <600 mm (Sheldon and Retallack, 2004)
fluvial depositional system, open woodland habitat, warm temperate climate with pronounced dry season, average winter temperatures probably not <13°C as indicated by giant tortoises (Evanoff et al., 1992)
Fluvial depositional system, open woodland habitat, warm temperate climate with pronounced dry season, average winter temperatures probably not <13°C as indicated by giant tortoises (Evanoff et al., 1992)
Subtropical woodland habitat with seasonal precipitation (450 mm/year), about 16.5°C MAT based on gastropods, average winter temperatures probably not <13°C based on giant tortoises (Evanoff et al., 1992)
annual rainfall >1,000 mm, MAT around 18°C, subtropical without freezing temperatures (Murphey and Evanoff, 2011)
The Modern Niche of Rhineura floridana
R. floridana displays an extremely restricted environmental niche with low probabilities (<0.3) of suitable conditions outside of the Florida Peninsula and parts of the Panhandle (Fig. 1a). The niche model is largely dependent on two variables: precipitation of the driest month and temperature seasonality, the latter describing the amount of temperature variation over a given year (or averaged years) based on the standard deviation of monthly averages temperatures (Fig. 1b,c; Table 2). Together these account for 82.5% of variable contributions to the final (averaged) model. AUC values, describing changes in model performance between runs with and without test data, were also very high (>99%), indicating an almost perfect distinction between presence and absence data given the environmental variables. General environmental tolerances of R. floridana indicate warm, humid niche preferences, with rainfall of 1,161–1,471 mm a year and mean annual temperature (MAT) ranging from 19.3°C to 22.6°C (Table 2). The lowest monthly temperatures experienced by R. floridana fall between 3.9°C and 10.7°C, well above the freezing point for the fossorial environment in which they live.
Table 2. Relative contributions of environmental variables to the species niche model and variable range values for extant Rhineura floridana
R. floridana range
Variables without units are dimensionless indices.
Precipitation of driest month
Temperature seasonality (standard deviation)
Mean annual temperature
Precipitation seasonality (coefficient of variation)
Although soil ranked relatively low as a predictor variable, modern R. floridana inhabits only five dominant soil types with restricted distributions in North America: dystric regosols (61% of occurrences), gleyic podzols (18%), ferric acrisols (18%), gleyic luvisols (3%), and the single Georgia record on gleyic acrisols (Fig. 1d). Each is characterized by a high proportion (>60%) of sand and little silt or clay. All are medium and coarse in texture. Most of these soil types only marginally leave Florida or have isolated patches outside of the state, such as gleyic acrisols in Georgia and gleyic podzols skimming the eastern coasts of North and South Carolina.
Extinct Rhineurid Relationships
The combined phylogenetic analysis resulted in a sister taxon position of Rhineuridae relative to the remainder of crown Amphisbaenia (Fig. 2), congruent with other molecular phylogenies (e.g., Kearney and Stuart, 2004; Macey et al., 2004; Vidal et al., 2008). Oligodontosaurus wyomingensis, the second oldest known amphisbaenian from North America (late Paleocene), appears as a stem taxon outside of the crown clade. Oligodontosaurus was first described as a lizard by Gilmore (1942) and was later reassigned by Estes (1965) to its own family, Oligodontosauridae. Kearney (2003) placed Oligodontosaurus within Rhineuridae, primarily on the basis of a low tooth count (9 dentary teeth vs. 6–8 in most amphisbaenians). However, its position within Rhineuridae and even Amphisbaenia is still ambiguous, as it possesses features unique among either group (Berman, 1973). Cryptolacerta and Lacertidae form the consecutive outgroups.
Within Rhineuridae, relationships are largely similar to those obtained by Kearney (2003), although few nodes had high support (Fig. 2). For this reason, all nodes with posterior probabilities less than 0.7 were collapsed and a slightly incompletely resolved topology was used to illustrate paleoclimatic trends (Fig. 3).
Fossil rhineurids occupied a large range of habitats over the Cenozoic with varying climatic conditions, as indicated by local studies of isotopes, paleobotany, and paleosols (Table 1). Although the Eocene-Oligocene boundary marks a dramatic transition from greenhouse to icehouse conditions in North America (Zanazzi et al., 2007), paleoclimatic reconstructions for rhineurid-containing formations indicate more gradual localized changes. The earliest unambiguous records from the lower Eocene of Wyoming occurred in subtropical conditions with relatively high rainfall (>1,000 mm/year) and MAT around 18°C (Murphey and Evanoff, 2011). The middle and late Eocene witnessed increasing cooling and drying, culminating in subtropical woodland habitats with seasonal rainfall and relatively warm winters, as also indicated by the presence of giant tortoises at the same (White River Formation) locality (Evanoff et al., 1992).
Following the Eocene-Oligocene transition, rhineurids inhabited increasingly open environments such as wooded grasslands and savannas, with seasonal, temperate climates. Precipitation continued to decline into the late Oligocene, reaching steppe or even desert-like conditions in South Dakota (Sharps Formation; Prothero, 1994). Undiagnostic rhineurid remains from the early/middle Miocene of South Dakota and Nebraska (“R. sepultura” and “R. marslandensis”) indicate that rhineurids also persisted during this time in grassland habitats with fairly warm winters (Strömberg, 2002; Fox and Koch, 2004). The only valid rhineurid material found after the Paleogene belongs to the modern species R. floridana, known from various sites in Florida during glacial periods of the Pliocene and Pleistocene (Holman, 1958, 1959a; Meylan, 1982). Although local temperatures were significantly cooler in glacial times than they are there today (Willard et al., 2007), the presence of xeric-adapted fauna at two of the Florida sites indicates a relatively dry climate with well-drained soil (Meylan, 2005).
Environmental Tolerances of R. floridana
The current distribution of R. floridana is generally characterized by a warm and humid subtropical climate, with low temperature and rainfall seasonality and small diurnal temperature ranges (Table 2). According to the niche model, the species is highly restricted by two factors: precipitation during the driest month and temperature seasonality. Although the current range experiences high annual rainfall compared with the rest of the country, it receives relatively little precipitation during the driest period (Fig. 1b). Areas directly to the west are also wetter in the driest period, including parts of the Florida Panhandle and the Gulf Coast of Alabama, Mississippi and Louisiana. Variation in temperature seasonality shows roughly similar geographical patterns, with areas immediately to the north, south and west of R. floridana being more variable over the course of the year (Fig. 1c).
Although soil type was a surprisingly poor predictor of the current distribution, R. floridana generally appears limited to sandy, well-drained soils (Mulvaney et al., 2005). Due to its fossorial habits, the species is intolerant of overly wet conditions and will come to the surface following heavy rains (Ashton and Ashton, 1985). This may further help to explain its restriction to central and northern Florida, as the Everglades ecosystem to the south is a marshy wetland with frequent flooding, and areas to the north are exceedingly wet and seasonal. Likewise, the option to expand further west is blocked by unsuitable conditions along the Florida Panhandle and northern Gulf Coast. Potentially suitable habitat to the south is also obstructed by the Gulf of Mexico, presumably posing a substantial geographical barrier to fossorial organisms.
When compared with modern R. floridana, fossil records of the same species indicate a historical ability to tolerate drier conditions. During times of glaciation, Florida was much larger as a result of lower sea surface levels (Russell et al., 2009) and also much drier, with land contact to the semi-arid regions of southwestern North America (Webb, 1991; Woodburne, 2010). Meylan (2005) reported the presence of numerous xeric-adapted reptile, frog, and mammal species concurrent with R. floridana at the Pliocene Inglis 1A site, indicating a coastal savanna habitat. Fossil sites containing R. floridana also experienced greater temperature seasonality, though with average temperatures still above freezing (DeSantis et al., 2009). This is also supported by the presence of herpetofauna in the late Pleistocene (Illonian) of Florida similar to what is found there today (Holman, 1959a). In general, the fossil history of R. floridana suggests that they once had broader environmental tolerances than they do now, with the ability to survive in cooler, drier climates. Therefore, their relictual distribution in Florida may be the result of a physical inability to migrate to cooler, drier areas further west or south, as opposed to phylogenetically-inherited constraints on adaptation.
Environmental Adaptability in Rhineuridae
Like R. floridana, the family Rhineuridae also displays a wide range of environmental tolerances over its evolutionary history as evidenced by information from the fossil record. In contrast to the humid-subtropical environment of the Eocene, Oligocene rhineurids lived in increasingly dry and cool climates typical of open woodlands and savannas. The persistence of the clade in northwestern North America after the Eocene-Oligocene transition also suggests that rhineurids were able to adapt to environmental changes. This is independently supported by the phylogeny, showing that several Oligocene species such as Hyporhina galbreathi, H. antiqua, Dyticonastis rensbergeri, Oligorhineura sternbergii, and Pseudorhineura minuta are closely related to Eocene representatives from the same area (Fig. 2).
With the exception of the earliest rhineurids from the early/middle Eocene (Jepsibaena minor, Ototriton solidus, and Spathorhynchus fossorium), no rhineurid taxa experienced conditions as warm and wet as the modern species. They also most likely never experienced prolonged freezing conditions, despite a generally cooler climate in the late Paleogene and Neogene relative to the Eocene and before. The only potential indication of freezing temperatures from the rhineurid fossil record comes from the John Day Formation in Oregon, containing Dyticonastis rensbergeri (Berman 1976; Dillhoff et al., 2009). Dillhoff et al. (2009) compared the local paleoclimate to that of the modern-day southern Chinese highlands, which may reach subzero temperatures in winter but is generally considered to be warm-temperate to subtropical. However, it should be noted that these low temperature estimates are largely based on data from leaf physiognomy, which have been criticized on several occasions to underestimate true paleotemperatures, including those for the Oligocene of Oregon (Kowalski and Dilcher, 2003; Peppe et al., 2011). As such, it remains unlikely that significant periods of freezing actually took place during time of deposition.
Rhineurids disappear from northwestern North America only after the middle Miocene, which may be related to rapid and permanent global cooling following the mid-Miocene Climatic Optimum. Mid-Miocene climatic changes are associated with a pulse of evolutionary turnover in North America, including the diversification of grazing mammals in conjunction with the development of grasslands (Flower and Kennett, 1994). In contrast to mammals, reptiles are ectothermic and their distributions are highly dependent on temperature (Böhme, 2003). Therefore, the onset of prolonged freezing temperatures in midwestern North America after the middle Miocene may have exceeded many rhineurids' physiological tolerances, potentially driving them to extinction.
Relevant Scales of PNC and the Importance of the Fossil Record
Based on data from the fossil record, we showed that rhineurid amphisbaenians exhibited wide environmental tolerances over their evolutionary history, though perhaps not wide enough to withstand overly wet soils and the onset of prolonged freezing in northwestern North America. This physiological limit of environmental tolerances could be interpreted as PNC for the entire family, but also characterizes much of Amphisbaenia, which mainly inhabits tropical and subtropical regions (except for the Mediterranean genus Blanus, which hibernates during winter (Saint Girons, 1953; Gans, 2005)). This calls into question the appropriate phylogenetic scale at which to test for PNC. For example, if the majority of amphisbaenians are restricted to warmer climates and well-drained soils, is it evidence for PNC at a higher taxonomic level?
The issue of phylogenetic scale was discussed by Wiens (2008) and Losos (2008), with the conclusion that appropriate tests of PNC must include related taxa occurring in habitats where the group under study is absent. Only then can one determine if failure to occupy those habitats is due to the inheritance of niche-related trait(s) not shared by the other group. In addition to the phylogenetic scope, our results also emphasize the importance of the temporal scale at which PNC is investigated. By including fossil data both at the phylogenetic and ecological levels, we were able to gain a more complete picture of the ancestral environmental niche than could be derived from modern ecology alone. Application of such evolutionary perspectives to investigations of niche-restricted taxa may help us to better determine the prevalence of PNC at multiple scales, and to consider its role in species' responses to future climate change (short term) and global patterns of diversification and species richness (long term).
We are grateful to Kenneth L. Krysko (Florida Museum of Natural History) and John Jensen (Georgia Department of Natural Resources) for georeferenced locality data of R. floridana. We also thank Martin Kirchner (Museum für Naturkunde Berlin) for help with species modeling and Chris Bell (University of Texas, Austin) and Jason J. Head (University of Nebraska, Lincoln) for providing access to important literature. Patricia Holroyd (University of California, Berkeley), Hans-Dieter Sues (Smithsonian Institution, Washington), and Frank Tillack (Museum für Naturkunde Berlin) generously provided access to fossil and extant rhineurid specimens.